Integrated target evaluation and clinical translation assessment
KIF18A in Ageing
Six internal documents assess KIF18A. Five of them say the oncology case stands. They divide on whether anything in it reaches ageing, and this page sets out where the division lies, what each side reads, and what it would cost to settle.
Prepared for Internal reviewSources 6 reportsDated 2026-08-13
Six internal reports, one question
1. Executive Summary
Should Insilico pursue a selective KIF18A inhibitor for ageing, alongside its oncology development?
Not on its own, and not yet. Every registered study of a KIF18A inhibitor is an oncology study, none of them recruits an older or a senescent population, and no experiment has tested whether blocking KIF18A extends life or health in any animal. The ageing case in these reports is an argument built out of cancer data, and the two documents that examined it most closely disagree about whether that argument can be carried across.
The oncology case is separate and stands. Five of the six reports say so, and the disagreement is about ageing alone.
The strongest direct evidence connecting KIF18A to an ageing tissue argues against inhibiting it: in the oocyte, KIF18A is protective, and both documents that reach the question say blocking it would make matters worse.
Nothing here forecloses the target. What it forecloses is opening a standalone ageing programme before the measurements that would justify one have been made.
Clinical translation assessment, page 1
10clinical programmes, all in cancerTen is the higher of the two counts in the pack, and the clinical landscape section sets it beside the lower one.
0trials in an ageing populationBoth clinical reports reach that count on their own, searching different trial databases, and print it as a headline.
1programme with an efficacy readoutVLS-1488, in a conference abstract the clinical evidence review states it did not re-verify. Every other programme is a registry record.
Clinical translation assessment, page 1
Two claims are being run together, and only one has human trials behind it
Two claims are being run together in these documents and they have different amounts of evidence behind them. The first is that a KIF18A inhibitor kills cells carrying the wrong number of chromosomes and spares normal ones; that claim has human trials behind it. The second is that killing such cells in an ageing body would slow ageing; that claim has no direct experiment behind it at all, in any species. The reports that grade the case highly grade the first claim and inherit its confidence for the second.
Clinical translation assessment, page 1
Ageing readouts written into oncology protocols still in draft
The cheapest measurements that would move this answer are measurements inside trials that are already running. Senescence, inflammatory secretion and aneuploidy readouts added to the oncology protocols now, while the protocols are still in draft, would turn a cancer study into evidence about ageing at close to no extra cost. The list is set out at the end of this page.
2. Target Biology and Genetic Tolerance of KIF18A Loss
KIF18A is a motor protein. It walks along the fibres a cell builds to pull its chromosomes apart, and it controls how fast the ends of those fibres grow and shrink. Losing it is survivable: mice without it live, and the human population carries loss-of-function variants without an obvious phenotype attached to them.
That tolerance is the entire commercial argument, and it is also the first place the pack’s evidence thins. The evaluation grades the genetic case on a constraint band it could not reach the underlying database to verify, and its safety read rests on the same mouse work that the ageing argument reads in the opposite direction.
7solved structuresBest resolution 2.2 ångström.
Nonegenome-wide association hitsThe evaluation looked and found nothing to report.
20variants called pathogenicOut of 152 recorded in all.
36.4%identity to the closest relativeKIF18B is the nearest of the family members the evaluation grades.
The evaluation records 7 solved structures and 152 variants for KIF18A, and prints its tolerance of loss as a band rather than as a value.
KIF18A
Protein
kinesin family member 18A
Family
kinesin-8
Motor domain
residues 11-355
UniProt
Q8NI77
NCBI Gene
81930
Solved structures
7, best 2.2 ångström
Tolerance of loss
LOEUF < 0.6
Variants recorded
152, of which 20 pathogenic
Association study hits
None recorded
Baseline expression, nTPM
testis 12.8, lymphoid tissue 8.9, bone marrow 7.8
Every figure in the card above is the indication prioritization report’s own, taken from its target characterization summary and from its sections 2.1, 7.1, 7.2 and 8. Its methodology table credits identity to UniProt and NCBI Gene, structures to PDBe and RCSB PDB, variants to GWAS Catalog and ClinVar, species identity to Ensembl Compara and baseline expression to the Human Protein Atlas, each accessed 2026-08-13. The constraint band is the one figure with no row in that table.
No GWAS association at any trait
Moderate and indirect. The evaluation grades the genetic support moderate and indirect, and is explicit about the kind of evidence it has: somatic rather than inherited. It found no genome-wide association study hit for KIF18A at all. What stands in place of one is overexpression across cancer types, a dependency confined to cancer cell lines carrying unstable chromosomes, and the 20 of 152 ClinVar variants it classifies as pathogenic. The constraint figure and the dependency data it leans on are both drawn from sources the report elsewhere says it did not have.
The verdict rests on somatic overexpression, a dependency confined to cell lines with unstable chromosomes, and the pathogenic share of the ClinVar count. No inherited variant supports it.
What it turns on
The evaluation’s own words
Where it says so
The verdict in the evaluation’s own words
Moderate/indirect genetic support. No GWAS hits, but strong somatic overexpression data (Section 5.2), CIN-selective dependency (DepMap/functional studies), and ClinVar pathogenic variants collectively validate KIF18A as a real target. The evidence character is functional/somatic rather than germline-genetic.
Selection against losing the gene, and an animal that loses it and lives
This is consistent with KIF18A being an essential mitotic gene — loss-of-function variants would be strongly selected against, limiting common-variant associations.
Claims the report undercuts elsewhere in its own text
A constraint band, and a note that the database was out of reach. Both sentences are the evaluation’s own. It prints the constraint band in section 7.2 and records in section 11 that it could not reach the database the band comes from. Neither sentence acknowledges the other, and this page prints both rather than choosing between them.
A dependency dataset cited as evidence, and listed among the checks not run. The second sentence heads a list of five orthogonal checks, and the fourth of them is “IMPC KO / DepMap (oncology)”. The evaluation therefore rests part of its genetic verdict on DepMap in section 7.2 and files DepMap in section 11 under work it did not do.
Selection against losing the gene, and an animal that loses it and lives. The first sentence explains the absence of association hits by strong selection against loss of function. The second records a mouse without the gene that survives to adulthood. The evaluation prints both, three sections apart, and reconciles them nowhere.
Two databases named as sources, and missing from the table of sources. The table in section 11 gives a database and an access date for every numbered section of the report. Neither gnomAD nor Open Targets has a row in it. gnomAD is named in the line quoted here and its constraint band is printed in section 7.2; Open Targets is named in that line and appears nowhere else in the report at all.
Male infertility in null mice, no dose-limiting toxicity to 800 mg
Predicted tolerable, with one confirmed reproductive finding. The evaluation expects inhibition to be tolerated by healthy tissue, and grounds that on preclinical work rather than on a clinical safety database. The one confirmed finding that follows from the target itself is reproductive: mice without the gene are infertile males and otherwise live. The risk it names as theoretical is bone marrow, which follows from where the protein is expressed in healthy people. The furthest clinical read-out it cites is a Phase 1 dose escalation with no dose-limiting toxicity up to 800 mg.
The safety case rests on preclinical work and one Phase 1 dose escalation, not on a clinical safety database.
What it turns on
The evaluation’s own words
What the evaluation predicts
KIF18A inhibition is predicted to be well-tolerated in normal cells based on preclinical data; Kif18a-null mice show male infertility (germinal cell aplasia) but are otherwise viable.
The therapeutic window it is relying on
KIF18A is a mitotic gene — inhibition is tolerated by normal cells but lethal to CIN-high tumors (synthetic lethal window).
The risk it names, and calls theoretical
Bone marrow toxicity is the theoretical risk given expression in proliferating hematopoietic cells.
Where the protein sits in healthy tissue
Expressed predominantly in mitotic/proliferating cells; tissue-level expression highest in testis (12.8 nTPM), lymphoid tissue (8.9 nTPM), and bone marrow (7.8 nTPM).
The furthest clinical read-out it cites
No dose-limiting toxicities observed up to 800 mg in VLS-1488 Phase 1 data (ASCO 2025).
Two consecutive sentences in one paragraph: the first classifies the gene as essential, the second builds the therapy on its being dispensable.
The clash
What the evaluation says
Essential, and dispensable, one sentence apart
KIF18A is classified as an essential protein and a predicted intracellular protein (HPA).
Essential, and dispensable, one sentence apart
The critical therapeutic insight is that KIF18A is dispensable for normal cell division but selectively required by CIN-high tumor cells, which depend on KIF18A to manage the consequences of elevated chromosome mis-segregation rates.
A therapeutic argument resting on a classification it contradicts
Essential, and dispensable, one sentence apart. The two sentences are consecutive in the same paragraph. The first is a database classification; the second is the therapeutic argument the rest of the evaluation is built on. The evaluation does not say which of them governs, and the same split runs through the document: section 7.2 repeats the classification, section 9 repeats the argument.
Identity falls from 96.8 per cent in macaque to 71.9 in rat, and the grade the evaluation assigns falls from excellent to moderate with it.
Species
Binomial
Identity, per cent
Model suitability
Macaque
M. mulatta
96.8
Excellent
Dog
C. familiaris
85.4
Good
Mouse
M. musculus
76.2
Moderate
Rat
R. norvegicus
71.9
Moderate
Mouse and rat sit below the evaluation’s own 80 per cent threshold
Mouse (76.2%) and rat (71.9%) identity is moderate, below the >80% high-confidence threshold.
Macaque (96.8%) and dog (85.4%) are preferred translational models.
Every relative the evaluation grades sits below 37 per cent identity to KIF18A, which is the whole of its case that selectivity can be designed for. The protein the same paragraph calls the key selectivity concern has no bar here, because the table this draws does not list it.
The five paralogs and their identities are the evaluation’s own table in section 8. Its methodology table credits that section to Ensembl Compara, accessed 2026-08-13. Each identity is to KIF18A, and the evaluation records no alignment method behind the figures.
Every clinical inhibitor reports selectivity over the closest paralog
Achievable. The evaluation reads the paralog risk as something a chemist can design around. Its closest family member sits at 36.4 per cent identity and every other one at 27.6 per cent or below, which it treats as divergence enough for a selective molecule, and it reports that every clinical inhibitor already claims selectivity over that closest paralog. The protein it actually warns about is not in the paralog table at all. It is KIF11, the other mitotic kinesin, where losing selectivity would bring back the bone marrow toxicity that stopped an earlier class of drugs.
The evaluation’s grounds for calling selectivity achievable, and, in its last sentence, the protein those grounds do not cover.
What it turns on
The evaluation’s own words
The closest relative, and how close
KIF18B is the closest paralog at 36.4% identity.
Why the evaluation thinks selectivity is reachable
While the motor domain is conserved across the kinesin family, the overall sequence divergence (all paralogs < 37%) suggests selectivity is achievable.
What the clinical compounds already report
All clinical KIF18A inhibitors report selectivity over KIF18B and other kinesin family members.
The protein the evaluation actually warns about
The key selectivity concern is KIF11 (Eg5), the other mitotic kinesin with clinical-stage inhibitors; KIF18A inhibitors must avoid KIF11 inhibition to prevent the bone marrow toxicity that limited Eg5/KSP inhibitors.
The key selectivity concern has no row in the paralog table
The case for selectivity rests on the five relatives in the figure above. The closest of them, KIF18B, shares 36.4 per cent of its sequence with KIF18A and the rest are further off still, which the evaluation reads as room enough for a molecule that hits one and not the others. It then names a different protein as the concern that actually matters, and that protein has no row in the table the figure draws.
So the evidence and the warning do not meet. The distances make selectivity a chemistry problem among the relatives the evaluation grades; the warning puts the risk that would cost the programme outside them, with another mitotic kinesin whose inhibition limited an earlier class of drugs. The evaluation prints both in one paragraph and reconciles them nowhere, and this page leaves them as it found them.
Structure
The motor domain, alone and on its microtubule
The interactive view could not be drawn in this browser, which needs scripting and a working graphics context to turn a structure. Every structure it would have shown is described below.
Tubulin dimerMagnesium ion marking a nucleotide pocketRest of the protein
The KIF18A motor domain alone, sharp enough to design against
The grey cartoon is the kinesin motor domain, the stretch of KIF18A the evaluation gives as residues 11-355 and judges suitable for structure-based drug design. The crystal holds two copies of the domain and the second copy is left out of this view. The purple sphere is a magnesium ion, drawn to say where the nucleotide pocket is: in the coordinates, adenosine diphosphate (ADP) sits beside it in that pocket, and the viewer leaves the nucleotide undrawn. No inhibitor is bound anywhere in the file, so what this entry offers a chemist is the druggable pocket the evaluation describes, the one that binds adenosine triphosphate (ATP), with the hydrolysis product still in it rather than a compound.
The grey cartoon is KIF18A. The olive pair beneath it is the piece of the microtubule the motor is holding: one alpha-tubulin and one beta-tubulin, drawn as a ribbon with thin sticks. Two purple spheres mark the nucleotide pockets in the drawn chains, both magnesium ions: one in the motor, where the file also holds adenosine diphosphate (ADP), and one in the alpha-tubulin, beside its guanosine triphosphate (GTP). The beta-tubulin carries guanosine diphosphate (GDP) and a molecule of taxol in the coordinates; none of the nucleotides, and not the taxol, is drawn. The deposited construct is a chimera of KIF18A with a methyltransferase, as the file’s own records say, and nothing of that fusion partner is resolved, so the chain on screen is KIF18A itself, with no inhibitor bound to it.
Seven experimental structures of KIF18A are deposited, and these are the two this page draws: the sharpest view of the motor domain on its own, and the motor seated on the microtubule it works on. Both files cover the same working part of the protein, and neither contains an inhibitor. Each holds adenosine diphosphate (ADP) with a magnesium ion where a nucleotide binds; the viewer draws each magnesium as a sphere to mark the pocket and leaves the nucleotides themselves undrawn. The sharpest of the seven is the file the evaluation calls suitable for structure-based drug design, and the pocket its sphere marks is the ATP-binding pocket the evaluation calls druggable.
Coordinates as deposited in the Protein Data Bank, trimmed of solvent and crystallographic bookkeeping records; no atom was altered. Both identifiers, and the resolutions printed with them, are rows of the evaluation’s structure table in its section 2.1. The five other entries that table lists, including the 2.41 ångström tubulin complex 9YMG, ship as table rows only.
Thirteen indications, two scoring runs
3. Disease Relevance Across Thirteen Ranked Indications
The prioritisation report scored KIF18A across a field of indications and reported the leading ones by therapeutic area. Its shortlist is dominated by cancers with high chromosomal instability, which is the population a KIF18A inhibitor is designed to kill.
Two age-related entries appear in that shortlist. Neither carries a measured fold change of its own, and the report attaches a reliability flag to both saying they should not be pursued on its own scores alone.
1,000indications the indication report scoredThe indication report analysed the top 50 in oncology and age-related areas.
13indications the ranking report tieredEleven cancers, and two age-related entries in the bottom tier.
0.520largest log2 fold change in the expression tableHepatocellular carcinoma, higher in tumour than in matched normal tissue.
0.121the one age-related fold change in itAlzheimer’s disease, the smallest positive value that table prints.
Hepatocellular carcinoma at 0.520, Alzheimer’s disease at 0.121
The gap is the finding. The largest fold change in the table is 0.520 in hepatocellular carcinoma. The only age-related disease in it is Alzheimer’s disease at 0.121, the smallest positive value the table prints, and the report attributes that value to dividing glial cells rather than to a KIF18A mechanism.
Each bar is one disease’s log2 fold change — the number of doublings that separates KIF18A’s level in diseased tissue from matched normal tissue, so a positive value means more of it in the disease. Three of the fourteen values are negative, all of them blood cancers.
Nothing in the chart matches this selection. The indication table below keeps every row listed.
The one age-related diseaseThe other thirteen diseases
KIF18A differential expression across the fourteen diseases the indication report measured, ordered by fold change. Twelve are cancers, one is metabolic syndrome, and one is a neurodegenerative disease. The neurodegenerative row carries the smallest positive fold change in the table.
Values, p-values and q-values as the indication report’s differential-expression table prints them; a q-value is the false-discovery-adjusted p-value, and hovering a bar shows both. Disease names are that table’s own, including its spelling of Hodgkins lymphoma. Indication prioritisation, 5.2 Differential expression
Glial proliferation rather than a KIF18A mechanism
The Alzheimer’s disease signal (logFC=0.12) is weak and likely driven by glial proliferation rather than a direct KIF18A-disease mechanism.
What the fold changes are measured against. In normal tissue KIF18A is a proliferation marker, not a broadly expressed gene: the three tissues that carry the most of it are the three that divide the most. The report lists testis at 12.8, lymphoid tissue at 8.9, bone marrow at 7.8 normalised transcripts per million (nTPM).
Every Go is a cancer, and the three No-go calls include both age-related entries. The table holds the thirteen indications the ranking report tiered and three more the indication report ranked on its own — papillary renal cell carcinoma, diabetes mellitus, thrombotic disease.
The table is arranged as the two runs, either side of the disease name. Tier, score and the first fold-change column are the ranking report’s 13-indication run; the second fold-change column, the rank and the call are the indication report’s 1,000-indication run. What each run prints in its own words, and what to check about a row before using it, follow in the second table rather than crowding the first.
Four tiers, and both age-related entries in the fourth
Sixteen indications, five graded Go — every one of them a cancer.
Indication
TierRanking report.
ScoreComposite, out of 10.
Fold change, tier tablelog2, from the ranking report.
Fold change, expression tablelog2, from the indication report; the small note names the narrower disease that table measures.
RankPandaOmics position out of 1,000 indications.
CallThe indication report’s verdict.
High-grade serous ovarian cancer
1
9.2
—The tier table prints “N/A*” and footnotes that this subtype has no separate expression category in PandaOmics, ovarian cancer being ranked 22 as a whole.
—
22
Go
Breast cancer, triple-negative and basal-like
1
8.5
0.611
0.433as invasive breast ductal carcinoma
3
Go
Colorectal cancer, chromosomal-instability-high
2
7.8
0.405
—
5
Go
Hepatocellular carcinoma
2
7.2
0.520
0.520
4
Go
Non-small cell lung cancer
2
6.8
—The tier table prints “N/A”.
—
24
Go
Gastric cancer
2
6.5
0.493
0.493as gastric carcinoma
74The tier table prints “N/A”. The indication report ranks gastric carcinoma 74.
Conditional
Osteosarcoma and Ewing sarcoma
3
6.0
0.371
0.371as Ewing sarcoma
—The tier table prints “N/A”.
—
Head and neck squamous cell carcinoma
3
5.8
0.393
0.393
—The tier table prints “N/A”.
—
Bladder cancer
3
5.5
—The tier table prints “N/A”.
—
48
—
Glioblastoma
3
5.2
—The tier table prints “N/A”.
—
10
Conditional
Clear cell renal cell carcinoma
3
4.8
0.209
—
19
—
Papillary renal cell carcinoma
—
—
—
—
19
Conditional
Age-related oocyte aneuploidy and reproductive ageing
4
4.5
—The tier table prints “N/A”.
—
8The tier table prints “8 (infertility)”, so the rank belongs to infertility rather than to reproductive ageing.
—
Neurodegenerative disease
4
3.5
0.121
0.121as Alzheimer’s disease
40
No-go
Diabetes mellitus
—
—
—
—
37
No-go
Thrombotic disease
—
—
—
—
30
No-go
A dash is a value the source does not print rather than a result that missed a threshold. Every figure is the run named in the column note, at the precision that run prints it.
The cells the two runs print in their own words, for the sixteen rows that carry one, and what to check before using the row. Eight rows carry a quoted cell from both runs.
Indication
Chromosomal instabilityAs the ranking report prints it.
Scorecard cellThe indication report’s per-indication differential-expression cell, as printed.
What to check
High-grade serous ovarian cancer
Very High (~100%)
Indirect
The highest-scoring indication in the pack carries no fold-change measurement of its own. Both reports rank it on chromosomal instability prevalence and trial activity instead.
Breast cancer, triple-negative and basal-like
Very High (~80% in TNBC)
Yes (0.43)
The two reports measure different slices of breast cancer and print different numbers: 0.611 for basal-like breast carcinoma in the ranking report, 0.433 for invasive breast ductal carcinoma in the indication report. Neither figure is wrong; they are not the same disease category.
Colorectal cancer, chromosomal-instability-high
High (~65–85%)
Yes (0.49)
The indication report’s differential-expression table has no colorectal row at all, yet its scorecard prints a fold change for one.
Hepatocellular carcinoma
High (~60–70%)
Yes (0.52)
The one indication where all three numbers agree across both reports and the scorecard.
Non-small cell lung cancer
High (~60–75%)
Yes (0.39)
Graded Go on expression that was validated somewhere else. The number attached to it belongs to head and neck squamous cell carcinoma.
Gastric cancer
High (CIN subtype ~50%)
Yes (0.49)
The most statistically significant result in the pack, at p = 4.60e-86 across 19 datasets, and the only tier 2 indication the ranking report could not assign a rank to.
Osteosarcoma and Ewing sarcoma
High (~70–90%)
—
The ranking report scores a pair of sarcomas together; the fold change underneath belongs to only one of them.
Head and neck squamous cell carcinoma
High (~60–80%)
—
The fold change in this row is the one the indication prioritisation’s executive summary and its scorecard both print against non-small cell lung cancer.
Bladder cancer
High (~60%)
—
—
Glioblastoma
Moderate (~40–60%)
Yes (0.64)
The scorecard prints the largest fold change anywhere in the pack for an indication that neither report’s expression table contains.
Clear cell renal cell carcinoma
Moderate (~40–50%)
—
The lowest-scoring oncology indication in the pack still outscores both age-related entries.
Papillary renal cell carcinoma
—
Indirect
The indication report puts rank 19 on the papillary subtype and the ranking report puts the same rank on the clear cell subtype. Both are carried.
Age-related oocyte aneuploidy and reproductive ageing
N/A (aneuploidy mechanism)
—
The strongest ageing connection in the pack, and the one that argues against the drug rather than for it: the ranking report states that a KIF18A inhibitor would be contraindicated here, because KIF18A function is protective against oocyte aneuploidy.
Neurodegenerative disease
N/A (aneuploidy mechanism)
Minimal (0.12)
The lowest composite score of the thirteen indications either report scored. Its fold change is the smallest positive value the differential-expression table prints, and the report reads it as glial proliferation rather than a KIF18A mechanism.
Diabetes mellitus
—
—
The second age-related entry, scored by the indication report only. Its whitespace drivers are the protein interaction network at 0.97 and matrix factorization at 0.89, which is the co-embedding the report says the ageing scores are made of.
Thrombotic disease
—
—
The third No-go, and the only one that is neither cancer nor age-related. It is kept so the page does not imply the two ageing calls stood alone.
The first two columns are reproduced in the source’s own words and keep its spelling. The last column is this page’s, and says what the rest of the pack does to the row beside it.
Ranks 40 and 37 of 1,000, and No-go on both
Two age-related indications clear the indication report’s filter. Neurodegenerative disease ranks 40 and diabetes mellitus ranks 37, and both are graded No-go. The report does not leave the grade to speak for itself: it says the scores come from network co-embedding rather than from direct evidence, which means they measure where KIF18A sits in a protein interaction graph and not what KIF18A does in an aged or diabetic tissue. The clinical translation assessment reaches the same place from its own data and puts a number on it, an attention score of 0.000 for ageing. The thirteen-indication ranking reaches it from a third direction and places both entries alone in its bottom tier, below every cancer it looked at.
The report files the finding under a Red reliability band, and the two sentences below are its own.
Network co-embedding rather than direct evidence
The aging-related indications in the filtered set (neurodegenerative disease #40, diabetes mellitus #37) have low PandaOmics scores driven primarily by network co-embedding rather than direct evidence.
Aging-related indications (neurodegenerative disease, diabetes) lack any direct evidence and should not be pursued based on PandaOmics scores alone.
Three No-go verdicts, two of them the age-related case.
Indication
RankOut of 1,000 indications.
RationaleAs the verdict table prints it.
Scorecard cell
Score driversThe strongest components behind the PandaOmics score.
Reading
Neurodegenerative disease
40
No direct evidence; aging signal unvalidated
Minimal (0.12)
—
The scorecard gives it a differential-expression cell and nothing else: no genetic evidence, no class precedent, no preclinical model. It is the only row in that table with three consecutive “None” cells.
Diabetes mellitus
37
No direct evidence; network-only signal
—
PPI (0.97), matrix factorization (0.89), pathway (0.73)
Its own drivers make the argument. The three scores that put it on the list are all network and pathway measures, which is what “network-only signal” means.
Thrombotic disease
30
No biological rationale for KIF18A inhibition
—
—
—
A dash is a cell the verdict table does not carry for that row. The thrombotic row is kept so the two age-related refusals do not read as the only ones. Indication prioritisation, 10.1 Verdict per indication
Attention score 0.000 for ageing
The clinical translation assessment queried ageing directly rather than reading it off an indication list, and every dimension that would show a real disease association came back at zero.
KIF18A is not one of the genes that change with age. The ageing differential-expression meta-analysis returned 2362 genes at a false-discovery threshold of q below 0.05 and KIF18A was not among them.
The clinical evidence review reached the same factual finding from the trial registry rather than from omics, and stated it in its opening line.
Five of nine dimensions at zero, and the two high scores read as mitotic-network position.
Dimension
ScoreOn the assessment’s 0 to 1 scale.
Its reading
Attention Score (the mandated disease-relevance metric)
HIGH — but reflects mitotic-hub connectivity, not aging biology
Pathways (iPanda)
0.571
Generic pathway co-membership
The last two rows are the co-embedding the indication report named, measured. They are the only above-baseline scores ageing produced, and the assessment attributes both to KIF18A’s position in the mitotic network rather than to anything about ageing. Clinical translation assessment, pages 5 to 6
Absent from the ageing gene set, absent from the top 50, and no clinical evidence
KIF18A not present in the DEG set
aging is not the queried target’s leading indication — it is not in the top 50 at all
There is currently NO direct human clinical evidence — of safety, efficacy, or robustness — for KIF18A inhibition as an anti-aging intervention.
Scored 3.5 and 4.5 against 9.2 for the leading cancer
The thirteen-indication ranking sorted the same two age-related entries into a tier of their own, scored 3.5 and 4.5 against 9.2 for the leading cancer, and stated what that means for the dual-purpose case in one sentence.
The strongest ageing link in the pack points the wrong way. The ranking report reads the reproductive-ageing evidence as showing that KIF18A protects eggs from aneuploidy, which makes inhibition the opposite of the intervention that evidence supports.
The one route left open is indirect, and the report that describes it grades its own case as speculative: aneuploidy drives cellular senescence, senescence is a hallmark of ageing, and KIF18A dysfunction promotes aneuploidy in dividing cells. Every step of that chain is about cells that divide, which is the objection the clinical translation assessment raises against the whole ageing thesis.
No entry in three ageing databases, no classical hallmark, and four clocks where KIF18A is not a top feature.
Database or metric
KIF18A status
Hallmarks of Aging (HOA) count
0 — Not associated with any classical hallmark
GenAge database
Not present
Geroprotector list
Not present
ClinicalTrials.gov aging studies
Not present
Druggable gene
Yes
Aging clocks
Present in 4 clocks (AltumAge methylation 2022, PASTA transcriptomics 2025, ZhangBLUP methylation 2019, Mammalian Life History 2024) but NOT among top features
Statuses as the ranking report prints them. The clocks row is the same shape of result as the network scores: present in the model, not driving it. Thirteen-indication ranking, Aging Relevance Assessment, KIF18A Is NOT a Classical Aging Target
Contraindicated for reproductive ageing, speculative for the rest
The dual-purpose oncology-aging narrative does not hold for a KIF18A inhibitor compound.
A KIF18A inhibitor would be contraindicated for reproductive aging, as KIF18A function is protective against oocyte aneuploidy. The aging relevance is mechanistic/biological rather than therapeutic.
Aging applications for a KIF18A inhibitor are exploratory and speculative at this stage.
Fifteen untapped indications, fourteen of them cancers
One age-related indication in a list of fourteen cancers, and the three scores that put it there are a protein interaction score, a matrix factorization score and a pathway score. The same report grades it No-go eight sections later.
The indication prioritisation’s own ranking, with every indication that already has an active KIF18A inhibitor trial removed. Fourteen of the fifteen rows are cancers. The exception is an age-related indication the same report grades No-go.
RankOut of 1,000 indications.
Indication
Area
Total score
Score drivers
13
lung cancer
Oncology
10.7
PPI (0.98), heterogeneous graph walk (0.95), GWAS sub-modules (0.91)
4
hepatocellular carcinoma
Oncology
10.6
PPI (0.94), network (0.94), attention (0.90)
10
glioblastoma multiforme
Rare/genetic
9.5
PPI (0.97), GWAS sub-modules (0.90), matrix factorization (0.90)
80
colorectal adenocarcinoma
Oncology
7.8
PPI (0.95), matrix factorization (0.89), network (0.88)
46
medulloblastoma
Oncology
7.6
PPI (0.99), network (0.93), pathway (0.91)
6
liver cancer
Oncology
7.2
impact factor (0.84), attention (0.80), PPI (0.78)
74
gastric carcinoma
Oncology
7.1
PPI (0.99), matrix factorization (0.89), network (0.89)
19
papillary renal cell carcinomaThe ranking report’s tier table puts this rank 19 on the clear cell subtype instead.
Oncology
6.8
PPI (0.95), GWAS sub-modules (0.90), network (0.83)
prostate adenocarcinomaProstate appears twice in this list, at ranks 26 and 33, with the same score; only the third driver differs.
Oncology
5.8
PPI (0.96), matrix factorization (0.90), GWAS sub-modules (0.75)
26
prostate carcinomaProstate appears twice in this list, at ranks 26 and 33, with the same score; only the third driver differs.
Oncology
5.8
PPI (0.96), matrix factorization (0.90), GWAS sub-modules (0.89)
37
diabetes mellitus
Endocrine/metabolic
5.2
PPI (0.97), matrix factorization (0.89), pathway (0.73)
82
esophageal carcinoma
Oncology
4.9
PPI (0.99), matrixfact (0.91), network (0.88)
92
chronic myelogenous leukemia
Oncology
4.5
PPI (0.96), matrix factorization (0.88), pathway (0.80)
61
astrocytoma
Oncology
4.2
PPI (0.97), matrixfact (0.90), network (0.89)
Names, areas and drivers as the untapped-opportunities table prints them, near-duplicates included. Indication prioritisation, 3.3 Untapped opportunities
Thirteen indications against a thousand
4. Indication Prioritisation Across Two Scoring Runs
Where both runs name the same disease, their orderings disagree, and neither document reconciles itself with the other.
Two documents in this pack ran the same scoring engine over KIF18A and reported different fields. One scored thirteen indications and sorted them into four priority tiers; the other scored a thousand indications across fourteen therapeutic areas and reported positions out of that. Neither reconciled its numbers with the other’s.
Where both name the same disease, both positions are carried below and neither is preferred. The chart draws the diseases that appear in both, and what it shows is that the two orderings disagree: the indication the tier table ranks highest sits twenty-second in the wider field, and the reproductive-ageing entry the tier table puts last sits eighth.
Figure 3
Falls under the composite rankingRises under the composite rankingHolds its position
The same diseases under both methods. Diseases at the top of the expression tiers sit in the hundreds once everything else known about the genes is counted.
Left, the tier tables of the indication report. Right, the composite rank out of 1,000 from the target evaluation, for whichever of the one genes the ranking placed the disease under, taking the higher position where it placed both. Each gridline on the right-hand scale marks ten times the position of the one before it, so the top of the field is spread out and the tail is compressed.
Every age-related indication the evaluation ranked, with its position out of 1,000.
Indication
KIF18A
Stronger gene
What drives it
thrombotic disease
#30
KIF18A
The best-ranked cardiovascular indication in the top 50. The evaluation grades it No-go and gives as its reason that there is no biological rationale for inhibiting KIF18A at all. It prints no measures behind the position and no scorecard row under it.
Inborn errors of metabolism
#34
KIF18A
The best-ranked endocrine and metabolic indication in the top 50, and a congenital class of disease counted inside the seven the evaluation calls age-related. It is the only indication on this table the evaluation never grades, and the only one it prints neither measures nor a scorecard row for.
diabetes mellitus
#37
KIF18A
Carried by protein-protein interaction (PPI), matrix factorization and pathway scores. The evaluation grades it No-go, on the ground that the signal is network-only and nothing direct sits under it.
neurodegenerative disease
#40
KIF18A
The only neurologic indication the evaluation grades. Its differential-expression score of 0.12 is the lowest on the scorecard, and the same row records no genetic evidence, no class precedent and no preclinical model. Graded No-go, and the evaluation records the signal behind it as unvalidated.
A dash means the gene did not rank the indication at all.
How to read those positions
A position here is a position in a ranking of 1,000 indications, not a measurement of KIF18A in the disease. The evaluation states where the two it discusses come from: “The aging-related indications in the filtered set (neurodegenerative disease #40, diabetes mellitus #37) have low PandaOmics scores driven primarily by network co-embedding rather than direct evidence.” Co-embedding places a gene near a disease because the genes around it are near that disease, so a gene nobody has measured in a disease can still take a position for it. Three of these four are graded No-go: neurodegenerative disease and diabetes mellitus for want of anything direct behind the network signal, thrombotic disease because the evaluation records no biological rationale for inhibition at all. The fourth is graded nowhere. Inborn errors of metabolism appears only as the head of its therapeutic area, a congenital class of disease counted inside the seven the evaluation calls age-related.
The indications the ranking puts first, before any age-related filter is applied, with their positions out of 1,000.
Indication
KIF18A
Measures carrying the score
hepatocellular carcinoma
#4
PPI, network, attention
liver cancer
#6
impact factor, attention, PPI
glioblastoma multiforme
#10
PPI, GWAS sub-modules, matrix factorization
lung cancer
#13
PPI, heterogeneous graph walk, GWAS sub-modules
papillary renal cell carcinoma
#19
PPI, GWAS sub-modules, network
prostate carcinoma
#26
PPI, matrix factorization, GWAS sub-modules
prostate adenocarcinoma
#33
PPI, matrix factorization, GWAS sub-modules
diabetes mellitus
#37
PPI, matrix factorization, pathway
The full list runs to fifteen; the eight highest-ranked are shown. Indication names are printed with the capitalization the evaluation gives them, which is not the capitalization the indication report uses for the same diseases.
attention and impact factor carry no weight in the evaluation’s own appraisal
Literature and attention scores (weight-0) are not informative for target validation.
PandaOmics scores are relative, not absolute measures of association strength.
The gap the evaluation states in its own words
“No published evidence links KIF18A directly to aging biology, cellular senescence, or neurodegeneration.”
What the evaluation says its own ranking cannot do
DE (Section 5.2) uses PandaOmics pre-computed logFC/p-value.
GWAS/mutation are association-level, not fine-mapped.
gnomAD constraint values were not directly accessible in this environment.
Single prioritization source; scores are relative within PandaOmics.
Group means approximate PandaOmics aggregates.
Aging-related indications lack direct mechanistic evidence for KIF18A.
A high rank indicates the target-disease pair is more prominent across the 23 scoring dimensions compared to other indications — it does not imply causation or clinical validity.
Scores are driven by the volume and recency of literature, omics co-association, and grant funding — none of which is equivalent to clinical evidence.
Aneuploidy, senescence, inflammation
5. Ageing Biology: Aneuploidy, Senescence and Inflammation
Every link in the proposed mechanism has literature behind it; the chain as a whole has never been tested end to end.
The ageing argument in this pack is a chain of four links. Ageing tissue accumulates cells with the wrong number of chromosomes; those cells turn senescent; senescent cells inflame the tissue around them; and a drug that killed the aneuploid ones selectively would break the chain at its first link.
Each link has literature behind it. What the chain does not have is an experiment that tests it end to end, and the report that builds it says so in four separate places.
Ageing biology review, pages 2 and 6
4hallmarks touched by KIF18AGenomic instability, cellular senescence, stem cell exhaustion and inflammaging. .
4arguments for a therapeutic windowEach argument answers the objection that healthy cells are affected too.
0lifespan experiments runThe review says so in four separate places: in its executive summary, in its longevity section, in the evidence heat map and in its overall assessment. No lifespan or healthspan experiment using a KIF18A inhibitor or a KIF18A genetic model has been published.
5age-related resources checkedHallmarks of Aging (HOA) targets database lists at least one.
The four hallmarks the review connects KIF18A to.
Hallmark
What the review claims
What the claim does not cover
Confidencereview’s own
Genomic instability
Chromosomal instability (CIN) and aneuploidy are classified as Hallmark #1 in the updated López-Otín framework, and the oncology work the review cites reports that KIF18A inhibitors selectively kill aneuploid cells while sparing normal diploid cells. That selective clearance is read as addressing the hallmark directly.
The selective-killing result is from a paper titled for aneuploid cancer cells. No aged normal tissue was tested, and no aged-tissue data is printed anywhere in the review.
5 of 5
Cellular senescence
Aneuploid cells enter senescence through proteostasis failure and mitochondrial dysfunction, then activate the senescence-associated secretory phenotype (SASP). Killing aneuploid cells before they become senescent is proposed as a preventive senolytic aimed at the upstream cause.
No KIF18A senescence measurement is printed. The only quantity in the chain is a 25 per cent median-lifespan extension from clearing p16-positive cells, which is a senescence number rather than a KIF18A one.
4 of 5
Stem cell exhaustion
Progeroid mice carrying a mitotic-checkpoint insufficiency lose stem cell function across tissues. KIF18A is held to be dispensable for normal stem cell division, so an inhibitor would clear aneuploid stem cells and leave healthy ones to replenish tissue.
Two links carry no citation: that aneuploid stem cells accumulate with age and reduce regenerative capacity, and that KIF18A is dispensable in normal stem cells. Both are asserted in the same paragraph.
3 of 5
Inflammaging
The chain the review draws runs from chromosomal instability to micronuclei, from micronuclei to DNA in the cytoplasm, from cytoplasmic DNA to activation of the cGAS-STING innate immune pathway, and from there to chronic inflammatory signalling.
Every study in this chain is cancer or immune biology. None measures inflammation in an aged animal, and none of the three involves KIF18A.
4 of 5
Confidence is the review’s own five-point rating, reproduced rather than recomputed. Ageing biology review, page 2
The four rows of the selectivity table, printed under the claim that KIF18A inhibitors have “an exceptional selectivity profile”.
The argument
As the review puts it
Where it stops
More than 100 cancer cell lines profiled
The finding cell records more than 40 per cent toxicity in chromosomally unstable cells against under 20 per cent in chromosomally stable and in normal cells. It is attributed to Phillips et al., Nat Commun, 2025.
Under 20 per cent toxicity in chromosomally stable and in normal cells is not zero. The executive summary describes this same table as toxicity confined to chromosomally unstable cells. No PMID is printed for the row.
KIF18A knockout mice
The knockout mouse is reported as viable and fertile, with no gross somatic abnormalities. It is attributed to Payton et al., Nat Cancer, 2024.
One section later the ageing biology review states that motor-domain variants raise oocyte aneuploidy prematurely in humans and in mice. Fertility and reproductive ageing are left unreconciled. No PMID is printed for the row.
Human bone marrow cells
Bone marrow cells are reported as “Minimally affected (unlike other anti-mitotics)”, attributed to that same 2024 paper.
Minimally affected is not quantified. No number, no assay, no dose and no PMID are printed for the row.
Mechanism of selectivity
The mechanism cell reads “Normal cells can divide without KIF18A; CIN cells cannot”, attributed to Cohen-Sharir et al., Nature, 2021 (PMID: 33505028).
The mechanism row is the only one of the four carrying a PMID. The study behind it is a cancer cell line screen; no aged tissue is tested in it.
The four arguments are numbered in the review; the order is preserved. Ageing biology review, page 6
What the review concedes about its own therapeutic window
The table closes by calling this profile superior to traditional anti-mitotic agents and “precisely the therapeutic window needed for a chronic or intermittent anti-aging treatment”. The numbers printed above that sentence are more than 40 per cent toxicity against under 20 per cent, which is a roughly two-fold separation.
No aged animal and no aged human tissue appears anywhere in this table.
Three of the four rows carry no PMID, and two of those three rest on the same single citation.
The nearest small-molecule precedent the review can name, one section earlier, used a kinesin potentiator that improves mitotic fidelity, which is the opposite pharmacological direction to inhibition.
Ageing biology review, page 6
Clock weights and tissue expression
6. Epigenetic Clock Weights and Age-Related Expression
The two direct measurements offered are internally inconsistent: the clocks disagree on sign, and the tissue analysis reads a rise and a fall as the same result.
Two kinds of measurement are offered as direct evidence that KIF18A is involved in human ageing: the weight that epigenetic ageing clocks give it, and how its expression changes with age across human tissues.
Both are read here as the report printed them, including where the report’s own numbers disagree with each other. The clock coefficients carry opposite signs on the same gene between two of the models, and the tissue analysis reads a rise with age in some tissues and a fall in others as supporting the same conclusion.
Ageing biology review, pages 3 to 4
4clocks include one of the genesAcross 5 separate entries.
4of those include KIF18AKIF18A is the only gene the review looked for.
+0.6686the review’s strongest single signalKIF18A in the mammalian life-history clock, ranked 9 of 227 within that clock.
803donors in the largest tissue sampleWhole Blood, measured for KIF18A. The smallest of the samples drawn here holds 193 donors.
Ageing biology review, pages 3 to 4
The review’s own summary of this evidence
The review reports “Together, these findings provide moderate-to-supportive evidence that KIF18A downregulation/inhibition aligns with anti-aging epigenetic signatures”. The five streams behind that count are listed below, with the direction each one points.
Ageing biology review, page 3
The five evidence streams the review’s verdict is counted from.
Clock or dataset
Gene
Coefficient
Rankwithin the clock
What the review reads into it
Mammalian life-history clockSupports
KIF18A
+0.6686
9 of 227
Tagged “SUPPORTS inhibition”. Higher methylation at CpG cg01203708 predicts later sexual maturity, which is treated as a proxy for slower ageing and longer lifespan across mammals, so suppressing KIF18A is read as mimicking the slower-maturing, longer-lived phenotype. The same clock is described as cross-species: it ranks species, not individuals and not treatments.
AltumAge methylation clockCaution
KIF18A
+0.0267
5243 of 20262
In a column headed Interpretation for Inhibition, this row gives no interpretation for inhibition. It states only that a positive coefficient in an age-prediction clock means the CpG, cg14927277, contributes to predicted older age. Read the way methylation is read elsewhere in the same table, that points against inhibition. It is not tagged either way.
PASTA transcriptomic age-shift clockSupports
KIF18A
−0.0000129
5303 of 8112
Tagged “SUPPORTS inhibition”. The cell reads the negative coefficient as higher KIF18A expression being associated with younger transcriptomic age, then concludes that inhibiting KIF18A could be age-neutral or could shift the transcriptomic clock. Lowering an expression the same cell says predicts youth is not an argument for lowering it.
REG transcriptomic age clockSupports
KIF18A
+0.0000162
7596 of 8112
Not tagged, but counted favourably. A positive coefficient means higher KIF18A expression predicts slightly older transcriptomic age, and inhibition is said to reduce that contribution. Same gene identifier and same authors as the row above, opposite sign.
ZhangBLUP methylation clockCaution
KIF18A
Mixed
—
The feature cell reads “11 CpG sites”, the coefficient and sign cells both read “Mixed”, and the rank cell reads “Varied”. The interpretation cell records multiple CpG sites with predominantly negative coefficients, tallied as 8 negative and 4 positive, and a net direction it calls predominantly negative.
Locators on this page are the numbered sections of the review itself. The source is an HTML extraction with no page markers, so its exact section headings are recorded in the pack notes instead. Four discrepancies are reproduced as printed rather than corrected. The review states three times that KIF18A appears in 4 ageing clocks; the summary table beneath prints the five rows listed here. The mammalian coefficient is printed as +0.6686 in that table and rounded to +0.669 in the section explaining it, which is one number at two precisions rather than two readings of it. The ZhangBLUP row names its feature as 11 CpG sites and tallies 8 negative against 4 positive, while the detailed table underneath it prints 14 rows, 10 negative and 4 positive — three incompatible counts of one set. PASTA and REG carry opposite signs on the identical gene identifier ENSG00000121621 from the same 2025 preprint, and both are counted in favour. One further overlap goes unremarked in the source: cg14927277 is both the AltumAge feature and row 13 of the ZhangBLUP table. Ageing biology review, page 3
Figure 4
Gene
With age
16 tissue measurements shown
No tissue measurements match this selection.
KIF18A
Bars run left or right of zero according to whether expression rises or falls with donor age, and are ordered by how strong the correlation is. Every correlation here is weak: the widest bar, ovary at −0.274, accounts for under eight per cent of the variation between donors.
The sentence printed under this table counts 17 significant tissues out of 49 tested, 10 decreasing with age and 7 increasing. The table itself prints 16 rows: 10 decreasing and 6 increasing. The extra increasing tissue is claimed but never named, so it cannot be drawn here. Reproduced as printed rather than corrected. Figure 1 of the source is referenced by caption only; its image did not survive extraction and the per-panel correlations its caption says are annotated appear nowhere in the text, so these bars are drawn from the table and not from that figure. Ageing biology review, page 4
Caution
Both directions are read as supporting inhibition. Where KIF18A falls with age, those tissues are said to be losing their aneuploid cells through natural mechanisms already; where KIF18A rises, those tissues are said to be accumulating unstable cells that raise KIF18A to survive, and to be therefore where an inhibitor is most relevant. No printed correlation in this table could have counted against the thesis.
Ageing biology review, page 4
The five curated fields the review reports for KIF18A, taken from the single sentence in which it reports them.
Resource checked
KIF18A
Hallmarks of Aging (HOA) targets database
Present, HOA_count = 0
ClinicalTrial_Gov
No flag
Publication
No flag
Geroprotector
No flag
GenAge
No flag
Checked by the review at the date it was produced. Ageing biology review, page 2
The same table, read two ways
The absence is read as a curation problem rather than an evidence problem: “This means KIF18A has not been formally curated as a canonical aging target. However, this reflects a gap in curation rather than a lack of evidence, as the mechanistic connections are strong.” No evidence is offered for the curation-gap assertion itself, and four hallmarks are graded between three and five stars on the strength of it.
The clinical review prints the same fields without the dismissal — “KIF18A-specific aging data: none in humans.” — recording HOA_count = 0, absence from the GenAge, geroprotector and ClinicalTrials.gov ageing lists, and presence in 4 ageing clocks but not as a top feature. It draws no curation-gap inference from any of it. Note also that the no-flag entry for ClinicalTrial_Gov sits five sections above a table of five NCT numbers; the review never states that the flag means ageing trials specifically, and never reconciles the two.
No lifespan experiment has been run with a KIF18A inhibitor. These are the experiments the review offers in its place.
The BubR1 mitotic checkpoint is called functionally analogous to KIF18A in controlling chromosome segregation fidelity. PMID 15208629.
Baker et al., Nat Cell Biol (2013)
BubR1 overexpression, which lowers aneuploidy
Extended healthy lifespan in mice
Cited in the overall assessment as genetic proof that reducing aneuploidy extends lifespan. PMID 23242215.
Baker et al., Nature (2011)
Clearance of p16-positive cells in BubR1 mice
Delayed ageing-associated disorders
Carries the step from aneuploidy to senescent-cell burden that the rest of the argument depends on. PMID 22048312.
Baker et al., Nature (2016)
Clearance of p16-positive cells in wild-type aged mice
About 25 per cent extension of median lifespan
The only lifespan figure printed anywhere in the review. It belongs to senescent-cell clearance, not to KIF18A. PMID 26840489.
Barroso-Vilares et al., EMBO Rep (2020)
Kinesin MCAK potentiator UMK57 in ageing fibroblasts
Reduced chromosomal instability and delayed cellular senescence
The nearest small-molecule precedent the review has, and the same sentence concedes that the compound is a potentiator improving mitotic fidelity. PMID 32134180.
Every lifespan figure in this table belongs to BubR1 or to p16-positive cell clearance; none of it is a KIF18A measurement. The last row runs in the opposite pharmacological direction to inhibition, and the same study is listed three sections earlier, in the senescence evidence chain, as small-molecule inhibition. No clinical precedent at all is printed from the mitotic-kinesin class KIF18A belongs to. Ageing biology review, page 5
Programme design and risk
7. Proposed Ageing Programme and Its Risks
Each proposal opens with an experiment nobody has run, which is the measure of how far the evidence currently reaches.
If the ageing case were to be pursued, the reports set out what pursuing it would look like, and the shape of the programme is the clearest statement of how far the evidence actually reaches.
Every proposal below begins with an experiment that has never been run. That is not a criticism of the proposals; it is the measurement of the gap they exist to close.
Ageing biology review, pages 5 and 7
The four risks the review lists against its own proposal, with the severity it gives each one in its own words. It grades none of them.
Risk
Severity
What the review proposes about it
No direct longevity or lifespan experiments using KIF18A inhibitors or KIF18A genetic models have been published to date.
Not rated
This represents the principal evidence gap for the anti-aging thesis.
This selectivity profile — sparing normal dividing cells including bone marrow — is superior to traditional anti-mitotic agents and is precisely the therapeutic window needed for a chronic or intermittent anti-aging treatment.
While this used a kinesin potentiator (improving mitotic fidelity), it establishes the principle that modulating kinesin-mediated chromosome alignment affects cellular aging.
Not rated
Small-molecule inhibition of aging-associated CIN delays cellular senescence
KIF18A motor domain variants (e.g., T273A) prematurely increase oocyte aneuploidy in both humans and mice.
Not rated
Viable, fertile, no gross somatic abnormalities
Ageing biology review, page 5
How the last row is written
It is written as support. The reproductive finding is booked as a positive line item, graded four of five stars in the heat map, marked Yes for direction, and summarised as KIF18A variants accelerating oocyte ageing. What the finding shows, by the review’s description, is a 25-year-old homozygous carrier of a motor-domain variant at 45 per cent oocyte aneuploidy: reduced KIF18A function accelerating ageing in the tissue measured. The words contraindicated, risk and caution appear nowhere beside it, and the selectivity table one section later still describes the knockout mouse as fertile.
Ageing biology review, page 5
The four things the review proposes doing
By killing proliferating aneuploid cells before they can become senescent, KIF18A inhibitors could reduce the senescent cell burden — functioning as a preventive senolytic targeting the upstream cause (aneuploid proliferating cells) rather than established senescent cells.
The tissues where KIF18A increases with age (skin, blood, arteries) may represent sites where age-related CIN cells accumulate and upregulate KIF18A as a survival mechanism — making these tissues particularly relevant for an KIF18A inhibition strategy.
This suggests KIF18A inhibitors could selectively clear aneuploid stem cells while allowing healthy stem cells to replenish tissues.
The critical missing piece is a direct experiment testing KIF18A inhibitors in aging models (e.g., treating aged mice and measuring healthspan/lifespan endpoints).
Ageing biology review, page 7
What the third proposal assumes
It assumes KIF18A is dispensable in healthy stem cells. That is asserted in a single sentence — that knockout mice are viable with normal somatic cell function and that KIF18A is dispensable for normal stem cell division — with no citation printed beside it. The assertion sits against the reproductive finding printed three sections later, and against the third link of the very evidence chain it concludes, that aneuploid stem cells accumulate with age and reduce regenerative capacity, which also carries no citation.
Ageing biology review, page 7
Early-stage oncology, nothing else
8. Clinical Landscape and Trial Activity
No study in the pack enrols an ageing population or carries a geroscience endpoint.
Both clinical sources searched trial registries on 2026-08-13 and found the same thing: a crowded early-stage oncology field, and nothing at all outside it.
That is the whole of the human evidence. Everything this pack says about KIF18A and ageing rests on work done in patients with advanced cancer, and the two reports that looked hardest disagree about whether a bridge can be built from one to the other.
Clinical evidence review, page 1
10clinical assets retrievedThe source’s own count of records retrieved from its internal trials dataset. It does not list them one by one, so the note under the table sets it beside the nine identifiers this pack prints.
100%of them in oncologyEvery registered study enrols patients with advanced cancer, and the selection axis is chromosomal instability (CIN): the tumours these programmes recruit are the aneuploid ones. The count of ten is the clinical translation assessment’s, read on its page 1; the clinical evidence review counts at least nine, because it treats one compound with two registry entries as one asset.
0trials in age-related diseaseNot one registered study takes an ageing or a senescent population, and none sets a geroscience endpoint. The clinical evidence review and the clinical translation assessment reach that finding separately, from an internal trials dataset and ClinicalTrials.gov, and print it as a headline.
1with published clinical resultsOne programme has a reported efficacy readout: VLS-1488, in a conference abstract from 2025. The clinical evidence review received that abstract as an uploaded dossier and states that it did not go back to the primary literature for them. Every other programme in the table has a registry record and nothing else.
Clinical evidence review, page 1
The assessment’s own framing
“The entire KIF18A clinical enterprise is oncology, aimed at chromosomally unstable (CIN-high) tumors, and it is still early: the most advanced asset is Phase 1/2.”
Clinical translation assessment, page 1
Figure 5
Stage
9 programmes shown
No programmes match this selection.
Nothing in this selection has reached a trial, so there is nothing to plot. The programmes are listed in the table below.
In Phase 1 and 2In Phase 1b, Phase 1 or Phase 1a and 1b
Bar length is the number of registered studies retrieved for each programme, which is a measure of how much a sponsor has committed rather than of how well the drug works. Of the programmes drawn here, two are in Phase 1 and 2; one in Phase 1b; one in Phase 1a and 1b; four in Phase 1. HW-221043 is in the table but not in this figure: it was retrieved from internal records and carries no registry identifier to count.
Registry records, with compound status cross-checked against the Ageing biology review. Clinical evidence review, page 3
Nine programmes, counted as compounds rather than as registry records. The clinical evidence review counts at least nine distinct clinical assets; the clinical translation assessment prints ten rows, because GH-2616 holds two registry entries and appears twice. Where another report gives a different sponsor, identifier or stage, the difference is printed under the setting rather than resolved.
Programme
Sponsor
Stage
Setting
Registry entriesas retrieved
ATX-295 (ACNT-2)
Accent Therapeutics
Phase 1 and 2
Solid tumours including high-grade serous ovarian cancer, triple-negative and other breast cancer, and non-small-cell lung cancerRecruiting. One trial, four names. The clinical evidence review prints ATX-295 with ACNT-2 in brackets, the clinical translation assessment prints ATX-295 alone, the thirteen-indication ranking prints ACNT-2 alone, and the research directive states that the ranking miscalls it. The stage splits the same way: the clinical evidence review prints Phase 1 from ClinicalTrials.gov and Phase 1 to 2 from the internal dataset in one cell, Insilico’s indication prioritisation report prints Phase 1, and the clinical translation assessment prints Phase 1 and 2. The clinical evidence review adds FDA Fast Track in April 2025 and preclinical claims only.
Advanced solid tumours, with high-grade serous ovarian cancer, squamous non-small-cell lung cancer, triple-negative breast cancer, head and neck squamous cell carcinoma and carcinosarcoma namedRecruiting. The most advanced programme in the pack and the only one with a reported efficacy readout. All four reports that list it agree on the identifier and the stage.
High-grade serous ovarian, fallopian tube and primary peritoneal cancer, a dose-optimisation cohort with 120 patients recorded by the clinical translation assessmentActive and not recruiting on ClinicalTrials.gov, Closed in the internal dataset. The clinical evidence review discloses the status split itself and calls it the same state under two vocabularies, meaning no longer enrolling. The identifier is disputed: Insilico’s indication prioritisation report prints NCT06083416 for this programme, and the research directive says that identifier returns not found on ClinicalTrials.gov and asks for it to be corrected and flagged. The thirteen-indication ranking prints no identifier for sovilnesib at all.
Solid tumoursOne study open, one closed. The clinical evidence review gives the developer only as the internal dataset’s entry and names no company. Insilico’s indication prioritisation report names Genhouse and the research directive names Suzhou Genhouse Bio. The record count differs as well: the clinical evidence review and the clinical translation assessment both list two studies, NCT06329206 and NCT07260513, while Insilico’s indication prioritisation report and the research directive list only NCT07260513.
Advanced solid tumours, including breast, ovarian, endometrial and fallopian tube cancer, with 66 patients enrolledCompleted. The only completed study in the class and the only one with a final enrolment figure. It produced no published result: the clinical evidence review reports that the registry record returned design and primary outcome measures only, with no response rates, no tumour-shrinkage figures and no adverse-event counts. The research directive gives the sponsor as Volastra, formerly Amgen; the clinical translation assessment prints the transfer the same way.
Solid tumoursOpen. The sponsor is given three ways for one trial: Jiangsu Gensciences in the clinical evidence review, GenSci in Insilico’s indication prioritisation report, and Changchun GeneScience in the research directive. The compound name splits too, printed as GenSci-122 by the two clinical reports and as GenSci122 by the other two.
Non-small-cell lung cancer, ovarian cancer and other solid tumoursOpen and recruiting in China. The clinical evidence review also carries a Chinese registration number for this study, CTR20252636, which is not a registry format the identifier count on this page reads. Insilico’s indication prioritisation report omits this programme entirely; the research directive asks for it to be resolved.
Platinum-resistant ovarian cancerRecruiting. Insilico’s own KIF18A inhibitor, out-licensed. The clinical evidence review records the compound as Insilico Medicine in origin and prints its internal code beside the external name; house style keeps that code off this page. The clinical translation assessment names Menarini alone as sponsor, Insilico’s indication prioritisation report names Stemline alone, and the research directive names both.
Solid tumoursPlanned. Not yet registered. The clinical evidence review records an internal identifier and no registry entry; the clinical translation assessment prints that identifier as 701018 and flags it the same way. Neither Insilico’s indication prioritisation report, the thirteen-indication ranking nor the research directive’s verified table lists this programme, which is one reason their counts run lower.
none retrieved
Sorted by stage, then by how many registry records each programme has. The square marks the programme that has reached the furthest stage. Where the sources say something about a programme beyond its registry record, it is printed under the setting. Clinical evidence review, page 3
The record counts do not reconcile
The translation assessment reports ten records from its query and does not print them. This page can name nine: nine in the table above and no more for the anti-mitotic class’s healthy-volunteer studies, listed below. One of the programmes in the table carries no registry identifier at all, so it cannot be among the ten either. Nothing in the pack says which records the query returned, so both totals are left as the reports give them rather than adjusted until they agree.
Clinical evidence review, page 1
Healthy-volunteer studies do not test a benefit in later life
No healthy-volunteer study of a KIF18A inhibitor appears anywhere in this pack, and none for the older mitotic-motor agents the reports use as precedent. Every study named above enrols patients with advanced cancer, most of them heavily pre-treated.
An ageing indication would put the drug into people who are not ill. The clinical evidence review lists that as a standing risk: anti-ageing use implies chronic or intermittent dosing in non-cancer subjects, a far higher safety bar than late-line oncology, and no data address it. Tolerability in a fourth-line ovarian cancer population does not transfer to a well seventy-year-old.
Clinical evidence review, page 3
How far ahead the leader is
Two programmes share the leading stage, Phase 1 and 2: Volastra Therapeutics’s VLS-1488 and Accent Therapeutics’s ATX-295. The clinical translation assessment sets the ceiling in the same place and states it flatly: no Phase 3 study, no approval, and no efficacy readout in ageing anywhere in the class.
Clinical translation assessment, page 10
The one trial with published numbers
Everything below comes from NCT05902988, Volastra’s study of VLS-1488, reported in abstract 3012 at the American Society of Clinical Oncology (ASCO) 2025 meeting. It is the only place in this pack where one of these compounds has been given to people and the results written up in a journal.
One correction to the sentence above, and it matters: these figures were not written up in a journal. They come from a conference abstract that the clinical evidence review received as an uploaded dossier and did not independently re-verify. Two reports print the efficacy numbers, a third prints only the safety numbers, and the research directive asks for all of them to be adjudicated against the primary abstract.
What was measured
What was found
What it means for a long-term programme
Evaluable patients with tumour reduction
7 of 17, which the clinical evidence review renders as 41 per cent, in high-grade serous ovarian cancer
A single-arm fraction from a dose-finding study, not a controlled response rate. It says the compound does something in the tumour type carrying the most chromosomal instability.
Confirmed responses
3 partial responses
Three responses are the whole of the efficacy evidence for this mechanism in human beings.
Stable disease
6 patients
Counted separately from the tumour reductions by the clinical evidence review.
Internal arithmetic
3 partial responses plus 6 with stable disease is 9 patients, which is more than the 7 credited with tumour reduction out of 17 evaluable
Printed as the review printed it. Nothing in the pack reconciles the two counts, so nothing here adjusts them.
Dose-limiting toxicities
None at any dose level, in a heavily pre-treated, platinum-resistant population
Tolerability, which is what a first-in-human study is built to establish. It is not evidence of a wide therapeutic index, and the clinical evidence review says so in the same document.
Enrolment and dose
52 patients at doses up to 800 mg, per Insilico’s indication prioritisation report and the thirteen-indication ranking. The clinical translation assessment’s landscape table records enrolment for this study as not reported.
The two reports that print a number agree on it. The report this table is attributed to prints none, which is the reason the figures above carry a second locator.
Regulatory status
FDA Fast Track in October 2024, for platinum-resistant high-grade serous ovarian cancer
A Fast Track designation reflects unmet need and development pace. It is not a statement that the drug works.
The attribution names the clinical translation assessment and its landscape page, which is where this trial’s stage, sponsor, status and enrolment field were read. The efficacy and safety figures in the middle column are not in that report at all: they were read from the clinical evidence review, page 3, and from the thirteen-indication ranking, both of which attribute them to the same 2025 conference abstract. Clinical translation assessment, page 10
The population these numbers came from
The seventeen evaluable patients were heavily pre-treated and platinum-resistant, the population where a signal is most meaningful and least generalisable. The clinical evidence review calls the result preliminary proof of activity in oncology, and in the same paragraph says it is not yet controlled efficacy evidence and says nothing directly about ageing. It also lists the readout among its own risks, as immature and single-source.
Clinical evidence review, page 3
The expression ranking against the clinical record
The thirteen-indication ranking scored KIF18A across thirteen diseases, and the clinical evidence review set the top of that list against the clinical record. They agree about the ordering and disagree about what it is worth. An expression fold change and a Phase 1 readout are not the same class of evidence, and the ranking’s two ageing entries sit at the bottom of its own list.
Clinical evidence review, page 7
Thirteen indications, scored, set against what has actually been given to a person.
Indication
Size of the expression change
Significance as printed
Clinical evidence
Where that leaves it
High-grade serous ovarian cancer
Not scored by fold change, Tier 1, composite 9.2 of 10
Chromosomal instability near 100 per cent of tumours; PandaOmics rank 22
Phase 1 and 2, four programmes, one efficacy readout
The clinical lead, and the only indication where a person has responded
Triple-negative and basal-like breast cancer
0.611, Tier 1, composite 8.5 of 10
p = 4.77e-19 across 6 datasets
Named in the VLS-1488 and ATX-295 enrolment criteria; no readout
The strongest expression signal in the ranking, with no clinical result behind it yet
Hepatocellular carcinoma
0.520, Tier 2, composite 7.2 of 10
p = 7.08e-71 across 12 datasets
No KIF18A programme enrols it
Scored fourth by PandaOmics and untouched by the clinical field
Gastric cancer
0.493, Tier 2
p = 4.60e-86 across 19 datasets, the most statistically significant result across all diseases scored
No KIF18A programme enrols it
The largest statistical signal in the pack sits in an indication nobody is running
Chromosomally unstable colorectal cancer
0.405, Tier 2, composite 7.8 of 10
p = 3.44e-52; PandaOmics rank 5
No KIF18A programme enrols it
Third on the ranking, absent from the trial list
Oocyte aneuploidy and reproductive ageing
Not scored by fold change, Tier 4, composite 4.5 of 10
PandaOmics rank 8 for infertility
None, and none proposed
The one direct KIF18A-ageing link in the pack points the wrong way: both reports state that inhibition would worsen oocyte quality, so the drug is contraindicated here
Age-related neurodegeneration
0.121, Tier 4, composite 3.5 of 10
p = 0.01 across a meta-analysis of 21 datasets; PandaOmics rank 40
None
The ranking calls it not actionable for drug development at present, and Insilico’s indication prioritisation report marks it No-go
Composite scores, fold changes and p-values are the thirteen-indication ranking’s. The clinical column is the clinical evidence review’s account of what has reached a patient. Fold changes are combined meta-analysis values on a log scale; p-values are printed here in exponent notation rather than the ranking’s superscript form. Clinical evidence review, page 7
The comparison the assessment insists on
A fold change of 0.121 in Alzheimer’s disease, at p = 0.01 across 21 datasets, is the strongest ageing signal the ranking found, and it places that indication fortieth by rank and last of the thirteen it scored. The clinical evidence review arrives at the same place from the other direction: KIF18A has a human-ageing-association count of 0, is absent from GenAge, from the geroprotector list and from the ClinicalTrials.gov ageing lists, and is present in 4 ageing clocks but not as a top feature.
Clinical evidence review, page 7
What the clinical evidence review says would have to be true before each group of indications is worth funding.
Group
Indications
What would have to be true
Run now
High-grade serous ovarian cancer, triple-negative and basal-like breast cancer, chromosomally unstable colorectal cancer, hepatocellular carcinoma
Nothing further, but design differently. The review’s first recommendation is to position KIF18A as oncology-first, with the human safety and efficacy package coming from the chromosomally unstable cancer programmes and high-grade serous ovarian cancer in the lead. Its second is to select those populations prospectively and to fund a validated, method-robust companion diagnostic for chromosomal instability, which it calls the single biggest de-risking lever.
Bridge, do not launch
A dual-purpose oncology and geroscience positioning
Geroscience readouts embedded in the oncology trials: serial senescence, senescence-associated secretory phenotype and aneuploidy measurements in on-treatment biopsies, using the dasatinib and quercetin senolytic study as the design template. The clinical evidence review treats this as ageing-relevant human data at low incremental cost. The clinical translation assessment rejects the same positioning outright, so the bridge is contested inside the pack.
Do not open
A standalone anti-ageing programme, age-related neurodegeneration, reproductive ageing
Four things the clinical evidence review lists and the pack does not have: a controlled oncology readout showing durable single-agent efficacy under a validated instability score; a demonstrated marrow-sparing therapeutic window in human beings; a KIF18A-inhibitor healthspan or lifespan experiment, which it calls the missing keystone; and a study showing that an inhibitor reduces aneuploid or senescent cell burden in normal aged tissue. Reproductive ageing is excluded on mechanism instead: KIF18A protects the oocyte, so inhibiting it is contraindicated there whatever the other evidence says.
Clinical evidence review, page 8
The assessment’s decision, taken separately for each programme rather than for the target as a whole.
Programme
Call
On what grounds
Oncology in chromosomally unstable tumours, led by high-grade serous ovarian cancer
Go
Both clinical reports endorse it. The clinical evidence review makes it recommendation one and asks for prospective enrichment and a companion diagnostic. The clinical translation assessment marks the oncology route Recommended at HIGH confidence, and in the same sentence tells the reader not to open a geroscience programme beside it.
A dual-purpose oncology and ageing positioning
Conditional
The dual-purpose positioning is where the two clinical reports part company, and the split is left standing here. The clinical evidence review keeps the position, calling KIF18A oncology-first and ageing-bridged rather than standalone, and asking for geroscience measurements inside the cancer trials. The clinical translation assessment rejects it: it marks the dual-purpose framing Reject, calls marketing a single KIF18A inhibitor as both unsupported and mechanistically self-contradictory, and says this is the specific claim the other two reports reject correctly.
A standalone anti-ageing programme
No-go
The clinical translation assessment classifies the anti-ageing and dual-purpose indication as No-Go pending direct evidence that does not currently exist, at HIGH confidence, and records an evidence level of zero for ageing: no human clinical data and no direct preclinical KIF18A-ageing data. Its conditional option is to keep ageing as a watch rather than a programme, and it states that absent the triggers it lists the answer is No-Go. The clinical evidence review does not use the words and reaches the same finding: no direct human clinical evidence of safety, efficacy or robustness for KIF18A inhibition as an anti-ageing intervention.
Reproductive ageing with an inhibitor
No-go
The clinical evidence review’s final recommendation is not to pursue it, because KIF18A is protective in oocytes and inhibition is contraindicated there. The thirteen-indication ranking, which placed this indication above the other ageing entry at 4.5 out of 10, says the same in its own text: inhibition would likely worsen, not improve, oocyte quality. This is the one place where the strongest direct KIF18A-ageing evidence in the pack argues against the drug rather than for it.
Clinical evidence review, page 8
Five things nobody has measured
The clinical evidence review lists what is missing twice, once as risks and once as what would change its verdict. These are the items on both lists that no experiment anywhere in the pack has run.
A lifespan or healthspan experiment. No KIF18A inhibitor has been given to an aged animal to see whether it lives longer or better. The review calls this the missing keystone and asks for a direct aged-mouse study to be commissioned. Without it, in its own assessment, the anti-ageing thesis cannot mature past mechanistically plausible.
Chronic dosing in anyone who is not ill. An ageing indication implies chronic or intermittent dosing in people without cancer, which the review calls a far higher safety bar than late-line oncology, and it states plainly that no data address it. Every study in the table doses patients with advanced disease, most of them heavily pre-treated.
A marker of aneuploid-cell burden in normal aged tissue. The ageing case needs a way to measure how many cells in an old tissue are aneuploid. The review states that no such clinical assay exists, and calls this the crux for any dual-purpose use.
A locked threshold for chromosomal instability. No companion-diagnostic cutoff for KIF18A response has been reported in human beings. Today’s trials enrich mostly by tumour histology, using ovarian cancer as a proxy for near-universal instability, rather than by a validated score. The review adds two reasons this matters: instability measurements are method-dependent, and instability failed as a predictor of immune-checkpoint response.
A human comparison of marrow sparing. Sparing the bone marrow is the differentiating safety claim against the older anti-mitotics, whose canonical dose-limiting toxicity is marrow suppression. The review flags this itself: no quantitative human comparison was confirmed, and the claim should be treated as a preclinically-rationalised hypothesis pending mature clinical safety readouts, not as established human fact.
Clinical evidence review, page 7
How much weight the assessment puts on each of its own conclusions.
Claim
Confidence
On what basis
KIF18A inhibitors are being tested in human beings and are tolerated enough to advance
High
The review answers Yes, citing at least six distinct programmes and one completed Amgen Phase 1 study, NCT04293094. It grades its own confidence High. This is the only claim in the table it accepts outright.
There is early single-agent anti-tumour activity in chromosomally unstable tumours, in high-grade serous ovarian cancer
Low
The review answers Emerging and immature: the signals come from the uploaded ASCO 2025 dossier and are not yet mature tool-verifiable readouts. It grades its own confidence Low to Moderate, which this page draws at the lower end.
The mechanism has a built-in patient-selection axis, chromosomal instability and aneuploidy, that de-risks it against the older anti-mitotics
Moderate
The review answers Plausible, names this the key differentiator, and immediately notes that a locked companion-diagnostic cutoff for chromosomal instability is not yet established. It grades its own confidence Moderate.
An aneuploidy-selective senolytic works in human beings for ageing
Low
The review answers Unproven, with zero human data, and says the closest precedent is a different senolytic class, dasatinib with quercetin. It grades its own confidence Very Low, a step below the lowest this page draws.
Clinical evidence review, page 2
Structures and paralog selectivity
9. Structural Druggability and Kinesin Selectivity
The pocket the field aims at is one every kinesin has, and the evaluation’s summary and its table disagree about how far apart the closest relatives are.
Seven experimental structures of KIF18A are deposited and the sharpest of them is judged suitable for designing against. The pocket the inhibitors in this field aim at is the one that binds the cell’s fuel molecule, which is the pocket every kinesin has.
That shared pocket is where the selectivity question lives. The family has more than forty members, five of them close enough to KIF18A to be worth naming, and the evaluation’s own reading of how far apart they are does not agree with itself between its summary and its table.
The seven structures the evaluation lists one by one. Resolution is how finely the structure was resolved, measured in ångström, where a smaller number is a sharper picture.
Structure
Protein
How it was solved
Resolution, ångström
Share of the protein
3LRE
KIF18A
X-ray
2.2
—
9YMG
KIF18A
X-ray
2.41
—
9DI0
KIF18A
Cryo-EM
3.1
—
5OGC
KIF18A
Cryo-EM
4.8
—
7RSI
KIF18A
Cryo-EM
4.9
—
5OCU
KIF18A
Cryo-EM
5.2
—
5OAM
KIF18A
Cryo-EM
5.5
—
KIF18A: seven listed, by X-ray crystallography and cryo-electron microscopy.
What the KIF18A structures cover, and what has to be modelled
The 2.2 Å X-ray structure (3LRE) covers the motor domain and is suitable for structure-based drug design. The 2.41 Å structure (9YMG) captures KIF18A bound to a non-hydrolyzable ATP analog and tubulin, providing mechanistic insight.
The patent field
The evaluation names five holders with filings against this target. It calls the field active. Freedom to operate is the right to make and sell a compound without infringing a patent somebody else holds, and it is settled by the chemical series a programme runs on rather than by the protein it aims at.
The five patent holders the evaluation names. It lists no filings for any of them.
Holder
Filings
Chemical series
Volastra Therapeutics
—
—
Accent Therapeutics
—
—
Amgen
—
—
GenSci
—
—
Genhouse
—
—
The evaluation gives its whole patent read in one paragraph. It records about 693 patent documents against KIF18A, names the holders below, and prints no patent number and no priority date for any of them. Its methodology table credits that paragraph to FreePatentsOnline, accessed 2026-08-13. A dash means no chemical series was recorded for that holder, which is the case for five of five.
What the evaluation concludes about freedom to operate
The space is competitive but not yet crowded with approved composition-of-matter patents.
The competitive picture in the evaluation’s own words
Eg5/KSP (KIF11) inhibitors were the first mitotic kinesin inhibitor class to reach clinical trials.
Ispinesib (SB-715992) failed in Phase 2 across melanoma… with no objective responses in any trial.
The failure was attributed to: (1) no patient selection biomarker; (2) narrow therapeutic window (bone marrow toxicity); and (3) resistance via allosteric mechanisms…
KIF18A is a novel target with no approved drugs.
The most advanced program is VLS-1488 (Phase 1/2), with initial data presented at ASCO 2025 showing no dose-limiting toxicities up to 800 mg in 52 patients.
Tiers: Since no drugs are approved for any indication, all indications are Tier 2 (development-phase drugs exist) or Tier 3 (no drug against target for that specific indication). Tier 1 (approved drug) does not apply.
Three places the evaluation says a programme here could differentiate itself
KIF18A inhibitors are fundamentally differentiated by CIN-selective synthetic lethality — normal cells tolerate KIF18A loss while CIN-high tumors die.
The key differentiator from prior kinesin inhibitor failures (Eg5/KSP) is biomarker-guided patient selection using CIN status.
Combination with immune checkpoint inhibitors (anti-PD-1) in CRC is a high-value strategy based on recent preclinical evidence…
Mechanism and its citations
10. Mechanistic Rationale and Supporting Literature
Several citations do not support the claim they are attached to; they are marked here rather than removed.
What follows is the mechanistic case as the reports make it, with the supporting literature they cite, and with the places where a citation does not support the claim it is attached to marked rather than removed.
One example sets the pattern for the rest. A 2020 study is filed under evidence for inhibition, and the compound it used is not an inhibitor.
The evaluation makes nine mechanistic arguments and records five next steps, and it does not name the indications the same way in both: three of the headings appears word for word in the other list. They are set out below in the order the evaluation gives them rather than paired up.
The nine mechanistic arguments the evaluation makes, in the order it makes them.
KIF18A is dispensable for normal cell division but selectively required by CIN-high tumor cells. Marquis et al. (2021) demonstrated that CIN tumor cells specifically require KIF18A for proliferation. Payton et al. (2024) provided pharmacological proof-of-concept with AMG-650, showing selective killing of CIN-high cancer cell lines.
Breast cancer
KIF18A overexpression correlates with higher tumor grade and proliferative index in invasive breast cancer ( Kasahara et al. 2016 ). KIF18A is a predictive biomarker of poor benefit from endocrine therapy in early ER+ breast cancer ( Alfarsi et al. 2019 ), suggesting patients with high KIF18A tumors may benefit from KIF18A inhibition as an alternative strategy.
Hepatocellular carcinoma
KIF18A promotes cell proliferation and metastasis in HCC through functional assays ( Ren et al. 2024 ). Additionally, KIF18A induces EMT in hepatoma cells through the 5-LOX-dependent arachidonic acid pathway, providing a non-mitotic oncogenic mechanism ( Wang et al. 2025 ). HPA validates KIF18A as an unfavorable prognostic marker in HCC (TCGA + validation cohort, p=8.1e-7).
Colorectal cancer
Targeted deletion of Kif18a protects from colitis-associated colorectal tumors in mice through impairing Akt phosphorylation ( Zhu et al. 2013 ). Critically, targeting KIF18A triggers antitumor immunity and enhances PD-1 blockade in CRC with CIN phenotype ( Liu et al. 2025 ), supporting a combination strategy.
Ovarian cancer
The KIF18A inhibitor ATX020 induces mitotic arrest and DNA damage in CIN-unstable HGSOC cells ( Nair et al. 2025 ). Novel cyclohexenyl KIF18A inhibitors show activity in ovarian cancer models ( Zhang et al. 2024 ). Ovarian HGSOC has the highest CIN burden of any solid tumor, making it the primary indication for CIN-selective therapies.
Lung cancer
KIF18A overexpression correlates with poor prognosis in primary lung adenocarcinoma ( Li et al. 2019 ) and contributes to proliferation, migration, and invasion ( Chen & Zhong 2019 ). HPA validates KIF18A as a potential prognostic marker in LUAD (TCGA, p=2.2e-5).
Glioblastoma
KIF18A is identified as a potential therapeutic target in glioblastoma stem cells ( Stangeland et al. 2015 ). KIF18A interacts with PPP1CA to promote malignant development of glioblastoma ( Yang et al. 2023 ).
Prostate cancer
KIF18A expression is associated with increased tumor stage and cell proliferation ( Zhang et al. 2019 ). Circ_CCNB2 knockdown sensitizes prostate cancer to radiation through the miR-30b-5p/KIF18A axis ( Cai et al. 2022 ).
Aging-related indications
No published studies directly link KIF18A to aging, cellular senescence, or neurodegenerative disease. The PandaOmics signals in neurologic and metabolic disease areas are driven by network co-embedding (PPI/matrix factorization) rather than direct evidence. KIF18A’s role in genomic stability is theoretically relevant to aging (CIN and aneuploidy accumulate with age), but this remains unvalidated.
Every sentence in the right-hand column is the evaluation’s own, citations included. None of it was checked against the papers it names.
The five indications the evaluation carries forward, with the verdict it records against each.
Indication
Verdict
What would have to happen next
What would rule it out
Hepatocellular carcinoma
Go
Test KIF18A inhibitor efficacy in CIN-stratified HCC PDX models; validate CIN status as a biomarker in HCC patient cohorts.
Low CIN burden in HCC subtypes would negate the synthetic lethal mechanism.
Colorectal cancer
Go
Combination study of KIF18A inhibitor + anti-PD-1 in CIN-high CRC models…
If CIN-high CRC subset is too small… the addressable market may not justify development.
Glioblastoma
Conditional
Assess BBB penetration of current KIF18A inhibitors; test in orthotopic GBM xenograft with CIN-high cells.
Insufficient BBB penetration would require reformulation or intrathecal delivery, fundamentally changing the development path.
Gastric carcinoma
Conditional
Validate KIF18A overexpression and CIN status in gastric cancer patient cohorts (Asian populations).
If KIF18A expression does not correlate with CIN or prognosis in gastric cancer specifically.
Papillary renal cell carcinoma
Conditional
PheWAS on KIF18A LoF variants to check for kidney-related phenotypes; test in RCC cell lines stratified by CIN.
If papillary RCC has low CIN burden compared to clear-cell RCC.
Every one of these has a recorded result that would rule it out. Both columns carry the evaluation’s own words.
Four decisions, four cheap checks
11. Strategy Options and Recommended Path
Four positions are on the table in this pack, and the two clinical documents agree on three of them. They agree that the oncology programme should run, that a standalone anti-ageing programme should not, and that an inhibitor should not be pointed at reproductive ageing, where the evidence says the target is protective. They part company on the fourth, which is whether one compound can be developed and described as both at once.
That split is left standing below. Each position carries the reasoning of whichever document reached it, and where the two documents reached different conclusions both are printed.
Clinical translation assessment, page 8
Oncology in chromosomally unstable tumours, led by high-grade serous ovarian cancer
Both clinical reports endorse it. The clinical evidence review makes it recommendation one and asks for prospective enrichment and a companion diagnostic. The clinical translation assessment marks the oncology route Recommended at HIGH confidence, and in the same sentence tells the reader not to open a geroscience programme beside it.
Clinical translation assessment, page 8
A dual-purpose oncology and ageing positioning
The dual-purpose positioning is where the two clinical reports part company, and the split is left standing here. The clinical evidence review keeps the position, calling KIF18A oncology-first and ageing-bridged rather than standalone, and asking for geroscience measurements inside the cancer trials. The clinical translation assessment rejects it: it marks the dual-purpose framing Reject, calls marketing a single KIF18A inhibitor as both unsupported and mechanistically self-contradictory, and says this is the specific claim the other two reports reject correctly.
Clinical translation assessment, page 8
A standalone anti-ageing programme
The clinical translation assessment classifies the anti-ageing and dual-purpose indication as No-Go pending direct evidence that does not currently exist, at HIGH confidence, and records an evidence level of zero for ageing: no human clinical data and no direct preclinical KIF18A-ageing data. Its conditional option is to keep ageing as a watch rather than a programme, and it states that absent the triggers it lists the answer is No-Go. The clinical evidence review does not use the words and reaches the same finding: no direct human clinical evidence of safety, efficacy or robustness for KIF18A inhibition as an anti-ageing intervention.
Clinical translation assessment, page 8
Reproductive ageing with an inhibitor
The clinical evidence review’s final recommendation is not to pursue it, because KIF18A is protective in oocytes and inhibition is contraindicated there. The thirteen-indication ranking, which placed this indication above the other ageing entry at 4.5 out of 10, says the same in its own text: inhibition would likely worsen, not improve, oocyte quality. This is the one place where the strongest direct KIF18A-ageing evidence in the pack argues against the drug rather than for it.
Clinical translation assessment, page 8
Four open questions that need no experiment
Four of the eight open questions on this page can be closed by reading a document or writing a sentence into a protocol, rather than by running an experiment. They are listed in the order they should be done, cheapest first.
Read the ASCO abstract behind the clinical activity figure
One document, one denominator, and four reports either keep the sentence or lose it together.
State the rank threshold the ageing clock claim is applying
Both documents printed the rank and neither printed the cut.
Say which object the programme counts are counting
The directive’s table is already row-by-row checkable and settles it for named assets.
Add senescence, SASP and aneuploidy readouts to the oncology protocols now, while they are still in draft
It is the only step that turns a running trial into evidence about ageing.
How the six documents rank
12. Source Quality and Evidence Grading
The six documents were not written to one standard, and two of them do not survive being checked against their own tables.
The six documents in this pack were not written to the same standard, and reading them as though they were is the error this section exists to prevent. Two are assessments, three are arguments for a position, and one is a directive setting out what the others should cover.
The appraisal below is the pack’s own, extended with what reading the documents against each other showed. Where a document’s arithmetic does not survive being checked against its own tables, that is recorded here and itemised at the end of the page.
13indications appraised for KIF18AFive graded conditional, five graded go and three graded no-go.
Every indication the appraisal grades, thirteen in all, with what each grade rests on.
Indication
Protein
What it rests on
Grade
Why that grade
Breast cancer (TNBC)
KIF18A
Xenograft + PDX
Go
CIN-high, active Ph1/2 trials, biomarker (KIF18A/CIN)
Ovarian carcinoma (HGSOC)
KIF18A
HGSOC cell lines, PDX
Go
Highest CIN burden, 4 active trials, strong data
NSCLC
KIF18A
Xenograft
Go
Active Ph1/2, expression validated
Hepatocellular carcinoma
KIF18A
Cell lines, EMT model
Go
Strong expression/network, unfavorable prognostic, no trial
Colorectal cancer
KIF18A
Kif18a-KO mouse, PD-1 combo
Go
CIN-dependent, PD-1 combo potential, Kif18a-KO mouse data
Expression data, but CIN burden lower than top indications
Medulloblastoma
KIF18A
—
Conditional
Pediatric, high network score, limited evidence
Neurodegenerative disease
KIF18A
None
No-go
No direct evidence; aging signal unvalidated
Diabetes mellitus
KIF18A
—
No-go
No direct evidence; network-only signal
Thrombotic disease
KIF18A
—
No-go
No biological rationale for KIF18A inhibition
All thirteen grades come with a reason.
What the appraisal counts in the ranking’s favour, for KIF18A
KIF18A is not a hub gene — the mitotic/cell-cycle network is specific and biologically coherent. Pharmacological validation exists with multiple chemical series. Six clinical trials provide independent industry validation.
CIN-selective synthetic lethality is a well-characterized mechanism with direct functional evidence…
What the appraisal counts against it, for KIF18A
Red: Aging-related indications (neurodegenerative disease, diabetes) lack any direct evidence and should not be pursued based on PandaOmics scores alone.
Amber: No approved drugs yet — all programs are Phase 1/2. Efficacy in humans is unproven. The Eg5/KSP precedent is cautionary.
Network/PPI scores (weight-1) are high but correlated — they reflect KIF18A’s position in the mitotic network, not independent biological evidence.
The eleven retrievals the report says it made, and which part of the ranking each one carries.
What it was used for
Database
How it was retrieved
2, 3
PandaOmics (Insilico Medicine)
2026-08-13
2
UniProt, NCBI Gene
2026-08-13
2.1
PDBe / RCSB PDB
2026-08-13
4
ChEMBL, ClinicalTrials.gov (API v2)
2026-08-13
5.1
Human Protein Atlas (proteinatlas.org, CC-BY-SA 4.0)
2026-08-13
5.2
PandaOmics expression meta-analysis
2026-08-13
6
STRING v12.0, Reactome (CC-BY 4.0)
2026-08-13
7
GWAS Catalog, ClinVar (NCBI)
2026-08-13
8
Ensembl Compara
2026-08-13
9
PubMed
2026-08-13
IP
FreePatentsOnline
2026-08-13
Every retrieval records the date it was made.
What is still unsettled
13. Open Questions and Unresolved Disagreements
Twelve questions the six documents answer differently, with both positions and the place each was read.
Ageing biology reviewSupportsYes, on mechanism
“The evidence supporting KIF18A inhibition as an anti-aging strategy is moderate-to-strong at the mechanistic level but lacks direct experimental validation.”
7. Summary of Evidence, 7.2 Overall Assessment
Indication prioritisationOpposesNo
“Aging-related indications lack direct evidence and are not recommended for pursuit.”
10. Recommendation: Which Indications to Pursue, 10.1 Verdict per indication
Thirteen-indication rankingOpposesNo
“The dual-purpose oncology-aging narrative does not hold for a KIF18A inhibitor compound.”
Strategic Recommendations, 6. Aging Indications — Monitor Only
Clinical translation assessmentOpposesNo, with high confidence
“PandaOmics Attention Score 0.000; not an aging DEG.”
page 13
Its own caveat: the three source reports could not be opened this session; conclusions were re-derived from live data and match Reports 2 & 3.
Clinical evidence reviewCautionNot yet, and not never
“let the CIN-high cancer programs generate the safety/PD and biomarker package, then bridge to a geroscience biomarker study — rather than positioning it as a standalone anti-aging therapeutic today.”
page 2
Reading for this site
Four of the five say no in some form, and the one that says yes says it about the mechanism rather than about the evidence. That distinction is the whole disagreement. Nobody in the pack claims an ageing result exists; they differ on whether a mechanism with no result behind it is a reason to start or a reason to wait. The clinical translation assessment is the only document written to rule on this, and it ruled — but it also recorded that it could not open the documents it was ruling on, which is why its position carries that sentence rather than standing alone.
What settles it
One direct experiment: dose aged animals with a selective KIF18A inhibitor and measure healthspan or lifespan against control. Every document in the pack names its absence; none of them reports an attempt.
If unresolved
If the mechanistic reading survives, the ageing indication is an unfunded hypothesis worth an animal study. If it does not, the ageing framing is removed from the programme and the oncology case — which no document disputes — carries it alone.
Ageing biology reviewSupportsCited as support
“demonstrating that KIF18A dysfunction accelerates reproductive aging”
5. Longevity & Lifespan Experiments, 5.4 KIF18A and Reproductive Aging
Thirteen-indication rankingOpposesReads it the other way
“KIF18A inhibition would likely worsen, not improve, oocyte quality.”
Research directiveCautionNames it the decisive conflict
“Direction of effect — the load-bearing conflict.”
Core Objective, 1. Source Adjudication (complete before building on any report)
Reading for this site
Both documents cite the same paper and neither misreports it: motor-domain variants that impair KIF18A raise oocyte aneuploidy early. The ageing review states in its own sentence that dysfunction accelerates ageing, and then files the paper under evidence for a drug that induces dysfunction. Read once out loud, the reading is unopposed — no document in the pack defends the supporting interpretation after the direction is pointed out, and the ageing review never addresses it. That is why this is marked settled while the question above it is not. It would be overturned by evidence that the tissue-level effect of pharmacological inhibition differs in sign from the germline effect of a hypomorphic variant, which no document offers.
What settles it
It is already settled inside the pack. What remains is to say so on the face of the programme rather than leaving the paper cited on both sides of the argument.
If unresolved
The single strongest human genetic link between KIF18A and an ageing phenotype points away from an inhibitor. Removing it from the supporting column leaves the ageing case resting on cell-line selectivity and on an analogy to a different gene.
Ageing biology reviewSupportsIt acts before they arrest
“functioning as a preventive senolytic targeting the upstream cause (aneuploid proliferating cells) rather than established senescent cells”
2. Hallmarks of Aging Assessment, 2.3 Hallmark: Cellular Senescence
Clinical translation assessmentOpposesIt cannot reach them
“KIF18A acts only in mitosis. Aged tissue is dominated by post-mitotic and senescent (growth-arrested) cells that do not divide and express little KIF18A. An anti-mitotic agent cannot act on non-cycling cells — the very population the thesis wants to clear.”
page 7
Reading for this site
This is the sharpest pairing in the pack and the two documents are not talking past each other. The clinical assessment says the drug cannot reach the target population. The ageing review agrees and redefines the target population: not the arrested cells but the dividing ones that will become them. The redefinition is coherent and it is expensive. A preventive agent has to be given to people who are not yet ill, for years, at a dose that kills dividing cells — and the pack contains no chronic-tolerability data of any kind.
What settles it
Two measurements, both in aged tissue rather than in a cell line: what fraction of the aneuploid burden sits in cells still cycling, and whether inhibitor exposure lowers that burden in a living animal.
If unresolved
If the preventive framing holds, the ageing indication is a decades-long prevention trial, not a treatment. If it does not, the drug and the disease do not share a cell population and no dose reconciles them.
Ageing biology reviewSupportsLower it
“By clearing CIN cells, KIF18A inhibitors would reduce the source of chronic cGAS-STING-mediated inflammation — directly addressing the inflammaging hallmark.”
2. Hallmarks of Aging Assessment, 2.5 Hallmark: Altered Intercellular Communication (Inflammaging)
Clinical translation assessmentOpposesRaise it
“A KIF18A inhibitor would be expected to increase cytosolic DNA/STING tone — plausibly accelerating inflammaging rather than relieving it.”
page 7
Reading for this site
The two readings differ over timing, not over biology. Killing an unstable cell removes a long-term source of cytosolic DNA; the killing itself produces micronuclei and releases more of it first. Both documents are describing the same drug at different points on the same curve, and neither reports a measurement that tells you which term dominates in a living tissue. The pack also notes that what is wanted in oncology, where this signal drives antitumour immunity, is precisely what is not wanted in ageing.
What settles it
Serial inflammatory readouts in a dosed animal — interleukin-6, interleukin-8 and interferon-stimulated genes measured over weeks rather than at one timepoint, so that a transient rise can be told apart from a sustained one.
If unresolved
If the tone rises and stays raised, the drug worsens the hallmark the ageing case was built to address, and the mechanism argues against itself.
Ageing biology reviewSupportsA leading feature
“The aging clock data — KIF18A as a top-10 feature in the mammalian life-history clock with a coefficient direction consistent with slower aging”
7. Summary of Evidence, 7.2 Overall Assessment
Thirteen-indication rankingOpposesPresent but not leading
“Present in 4 clocks (AltumAge methylation 2022, PASTA transcriptomics 2025, ZhangBLUP methylation 2019, Mammalian Life History 2024) but NOT among top features”
Aging Relevance Assessment, KIF18A Is NOT a Classical Aging Target
Reading for this site
Both documents worked from the same four clocks and reached opposite descriptions of the same rank. The ageing review prints the rank it is describing in the table beneath the claim, so the disagreement is visible without leaving the pack. Which description is right depends on whether a rank inside the first ten of a few hundred features counts as leading, and the two documents never state the threshold they are applying. Membership of a clock is in any case a statement about correlation with age, not about causation, and neither document claims otherwise.
What settles it
State the threshold. A rank is only leading relative to a stated cut, and both documents omitted theirs.
If unresolved
Small on its own. It matters because the clock result is one of four pillars the ageing review lists in its closing summary, and a pillar that turns on an unstated threshold is not load-bearing.
Ageing biology reviewSupportsQuoted as a clinical result
1. Disease Indications with KIF18A as Target (Clinical Phase), 1.2 Key Clinical Results
Clinical evidence reviewCautionNot retrievable this session
“No response rates, tumor-shrinkage figures, or adverse-event counts were returned”
page 3
Reading for this site
The clinical evidence review carried the figure forward and said in its opening note that efficacy figures originating from the uploaded dossiers were attributed rather than independently checked. So the most-quoted number in the pack traces back to one conference abstract, cited by one document, and every later appearance is a copy of that citation. That does not make it wrong. It means the pack contains one instance of it, not four.
What settles it
Pull the abstract and read the denominator. Every document that reproduces the figure names the same abstract number, so there is one document to check.
If unresolved
The figure is the pack’s only human activity signal. If it holds, the oncology case is unaffected either way, because the oncology case does not rest on it. If it does not hold, four documents lose the same sentence at once.
Clinical evidence reviewCautionSix or more distinct
“Yes — ≥6 distinct programs, one Amgen Phase I completed (NCT04293094)”
page 2
Research directiveCautionSix verified assets
The directive prints a verified table of named assets with sponsor, registry identifier and phase, and it has six rows.
Core Objective, 2. Human Clinical Trial Landscape Assessment
Reading for this site
The three counts are compatible once you notice they count different objects: registered trial records, distinct development programmes, and named assets with a resolving identifier. None of the documents says which it is counting, so the numbers read as a contradiction when they are a units problem. It matters only because the counts are quoted in support of how crowded the field is, and crowdedness is an argument about programmes rather than about records.
What settles it
Count one thing and name it. The directive’s table already does this for named assets and is the only one of the three that can be checked row by row.
If unresolved
None for the ageing question. It bears on the competitive read, where the difference between six programmes and ten changes whether an in-house compound is entering a contested field or a crowded one.
Clinical evidence reviewCautionQueue it behind oncology
“let the CIN-high cancer programs generate the safety/PD and biomarker package, then bridge to a geroscience biomarker study — rather than positioning it as a standalone anti-aging therapeutic today.”
page 2
Thirteen-indication rankingOpposesStop, and watch
“Action: No active therapeutic development for aging indications. Monitor emerging literature on the aneuploidy-senescence axis and KIF18A’s role in aging clocks.”
Strategic Recommendations, 6. Aging Indications — Monitor Only
Indication prioritisationOpposesStop
“low PandaOmics scores driven primarily by network co-embedding rather than direct evidence”
This is the only axis with a decision attached, and the two answers cost very different amounts. Queueing is close to free: the oncology trials are running anyway, and adding senescence and aneuploidy readouts to them buys the ageing hypothesis its first human data at the price of a few assays. Stopping is also close to free, and it avoids attaching an ageing label to a programme whose own evidence does not carry one. What neither document recommends is the expensive option — a standalone ageing study — and on that they agree.
What settles it
Nothing external. This is a resourcing decision, and the pack has already supplied both readings of the same facts.
If unresolved
Choosing to queue means writing geroscience readouts into oncology protocols now, while they are still being written. Choosing to stop means the ageing claim comes off the programme description, which is the part that has been travelling furthest from the evidence.
It assembles a mechanistic chain — ageing accumulates aneuploid cells, aneuploid cells turn senescent and inflame the tissue around them, a KIF18A inhibitor kills aneuploid cells and spares diploid ones — and grades the resulting case moderate-to-strong. The chain is coherent and the report is honest about its largest hole, saying in four separate places that no direct lifespan experiment has ever been run. What does not hold is the supporting arithmetic: three of its headline counts disagree with the tables printed beneath them, its two ageing-clock coefficients carry opposite signs on the same gene, and its expression analysis reads both a rise and a fall with age as evidence for the same conclusion.
Indication prioritisationOpposesRanks fifty indications, flags the two age-related ones No-go, and says the ageing signal is the network talking to itself.
“Aging-related diseases: No published evidence links KIF18A directly to aging biology, cellular senescence, or neurodegeneration.” 1. Indication Prioritization, untapped opportunities
It runs the target through a filtered ranking of fifty indications and finds oncology occupying most of them. The two age-related entries that clear the filter — neurodegenerative disease and diabetes mellitus — are both flagged No-go, and the report gives its reason rather than leaving the flag to speak: their scores come from network co-embedding rather than from any measurement of KIF18A in those diseases. It also states in its own words that no published evidence links the target to ageing biology, senescence or neurodegeneration at all.
Thirteen-indication rankingOpposesScores thirteen indications, puts both ageing entries in the bottom tier, and says the dual-purpose story does not hold.
“The dual-purpose oncology-aging narrative does not hold for a KIF18A inhibitor compound.” Strategic Recommendations, 6. Aging Indications — Monitor Only
It scores thirteen indications on a composite scale and sorts them into four tiers. Eleven are cancers; the two age-related entries sit alone in the bottom tier, below every oncology indication it looked at. Its reading of the reproductive-ageing evidence is the sharpest thing in the pack: the same paper the ageing review cites as support shows KIF18A protecting eggs from aneuploidy, which makes an inhibitor the wrong direction of intervention rather than an untested one.
Clinical translation assessmentOpposesAsked to adjudicate the pack’s disagreement, and did: the ageing case is No-Go, the oncology case stands.
“There is no human clinical evidence — of any phase — that a selective KIF18A inhibitor (or any drug with an identical or adjacent mechanism of action) is safe, effective or robust for anti-aging, age-related degenerative disease, or a dual-purpose (oncology + geroscience) indication.” page 1, Executive Summary
This is the only document in the pack written to settle the argument rather than to advance a position in it, and it settles it against the ageing thesis with stated high confidence. It reports the ageing omics signal as literally zero — an attention score of 0.000, a relevance score of 0.000, and absence from the 2,362 genes differentially expressed in ageing — and it finds every one of the ten registered programmes pointed at cancer. Its mechanistic objection is the one the rest of the pack talks around: a drug that acts in mitosis cannot reach a tissue whose defining feature is that its cells have stopped dividing.
Clinical evidence reviewCautionFinds the same absence of human ageing evidence and draws a different conclusion from it: sequence the programme rather than drop it.
“There is currently NO direct human clinical evidence — of safety, efficacy, or robustness — for KIF18A inhibition as an anti-aging intervention.” page 1
It reaches the pack’s factual consensus — no human evidence of any kind for the ageing indication — and then declines to convert that into a No-Go. Its argument is about order rather than merit: let the cancer trials generate the safety, pharmacodynamic and biomarker package, then bridge to a geroscience study, rather than positioning the asset as a standalone ageing therapeutic today. It is the only document that both concedes the evidence gap and keeps the dual-purpose route open, which is why it is read here as caution rather than support or opposition.
Research directiveCautionNot a finding but the commission behind two of them, and the only document that names the disagreement out loud.
“A well-evidenced negative answer is a fully acceptable and expected deliverable, and is more useful to us than a manufactured positive one.” Core Objective
It withholds a conclusion by design, because its job was to ask for one. What it contributes instead is the pack’s own map of itself: fourteen conflicts between the other reports, set out with which document says what, and seven factual corrections it asserts against them — including a trial identifier that does not resolve and a compound attributed to the wrong company. It also states the standard the answer was to be held to, which is the reason this page reads the way it does: a well-evidenced negative was named in advance as an acceptable and expected result.
What is not in dispute.
The point
Reports
What that agreement is worth
No clinical trial of a KIF18A inhibitor in ageing, senescence or any age-related degenerative disease exists anywhere.
5
Every document that looks at the trial landscape reports the same empty set, and none of them qualifies it. The disagreement is about what an empty set licenses: an untested hypothesis in one reading, an untestable one in another.
No direct experiment has ever tested whether a KIF18A inhibitor extends lifespan or healthspan in any animal.
2
Conceded by the report arguing for the ageing case, in the same paragraph that grades that case moderate-to-strong, and named by the clinical evidence review as the single most important gap in the dossier. The two documents agree on the fact and part company on what it costs the argument.
KIF18A is a credible oncology target in chromosomally unstable tumours, and the clinical programmes are real.
5
The pack does not divide over the target. It divides over one indication, and every document in it — including the two that reject the ageing case outright — endorses the cancer programme.
The class has reached Phase 1 and Phase 1/2 and no further: no Phase 2 efficacy readout, no Phase 3, no approval.
5
Agreed on the ceiling, and not on the floor: the documents count between six and ten programmes depending on whether unregistered and planned assets are admitted.
Counted by hand across the source documents.
The reports this page is built from.
Report
Produced by
What it covers
Length
Position
Ageing biology review
—
Ageing biology and the case for it. Ageing clocks, GTEx expression against age, hallmark mapping, senescence and SASP, the BubR1 lifespan literature, selectivity profiling, and a graded evidence heat map. Written as an argument for a position rather than an assessment of one.
Web page
argues for
Indication prioritisation
PandaOmics Agent (Insilico Medicine)
Target characterisation and indication ranking. Structures, constraint, dependency, variants, interactors, cross-species identity and normal-tissue expression, then a filtered top-fifty indication list with a traffic-light call on each. The version used here is v2; an earlier v1 ranked a thousand indications and placed the ageing entries far lower.
Web page
argues against
Thirteen-indication ranking
—
Thirteen indications — eleven oncology, two age-related — each scored zero to ten on chromosomal-instability prevalence, expression, ranking, clinical validation and unmet need. Carries its own differential-expression tables, a clinical development landscape and an explicit ageing-relevance assessment.
Web page
argues against
Clinical translation assessment
Insilico Medicine
Clinical translation, evaluated against the ageing indication directly. The registered-trial landscape, the adjacent-mechanism precedent, druggability scoring, on-target toxicity, the interaction network, and a section-by-section adjudication of where the three earlier reports disagree.
14 pages
argues against
Clinical evidence review
Insilico Medicine
The human trial landscape and what can be built on it. Nine or more clinical assets with registry identifiers, phase and status, the same-mechanism precedent, a biomarker and patient-selection strategy, and a translational assessment of the dual-purpose route.
9 pages
argues for, with conditions
Research directive
Insilico Medicine
A commissioning document, carrying the pack’s conflict register and its corrections. Names the three earlier reports, tabulates where they contradict each other and themselves, supplies corrected identifiers, and sets the scope of the two clinical reviews written in answer to it.
Web page
argues for, with conditions
Titles, dates and page counts are as they appear on the documents.
Reproduced, not repaired
15. Defects Found in the Source Reports
Every figure below is printed as its report prints it; the entry says what it disagrees with.
Eighteen points were logged while reading the six reports: eight errors, one disagreement between two reports and nine notes on how the source must be read. None of them is hidden here, and none of them is repaired.
Where a report’s number disagrees with its own table, both are printed and the disagreement is named. Where a citation does not support the claim attached to it, the citation stays and the mismatch is recorded beside it.
CorrectionThe scorecard grades nine of the eleven indications the verdict table calls, and two of the two it leaves out are among the report’s three No-go calls.scorecard-omits-calls
Where it appears
Insilico’s indication prioritisation report, 10.2 Confidence & reliability. The table headed “Per-indication scorecard:”, read against the verdict table at 10.1.
Why it cannot stand
The verdict table at 10.1 gives each of the eleven a rank, a tier, a direction, a modality, a call and a rationale. The scorecard at 10.2 drops Diabetes mellitus (No-go) and Thrombotic disease (No-go), and marks none of them as dropped. The reliability flag printed immediately above the scorecard reads: “Aging-related indications (neurodegenerative disease, diabetes) lack any direct evidence and should not be pursued based on PandaOmics scores alone.” Diabetes mellitus has no row in the table directly beneath that sentence. The scorecard is the only place in the report where the evidence behind a call is set out column by column, so the calls it omits are the ones a reader cannot audit.
How this site handles it
This page takes its calls from the verdict table rather than from the scorecard, and says where the scorecard is drawn on that it covers fewer indications. No row is added to the scorecard to close the gap, because the gap is the finding.
CorrectionThe scorecard prints “Yes (0.39)” in the differential-expression column for non-small cell lung cancer, and the executive summary cites a fold change of 0.39 at p = 2.99e-75 for the same indication.de-nsclc-0.39
Where it appears
Insilico’s indication prioritisation report, 10.2 Confidence & reliability. Executive summary and per-indication scorecard, against non-small cell lung cancer.
Why it cannot stand
The differential-expression table has no non-small cell lung cancer row. The values 0.393 and 2.99e-75 belong to head and neck squamous cell carcinoma, which is a separate row of that table. The tier table in the ranking report prints “N/A” for this indication’s fold change.
How this site handles it
Both figures are reproduced as the report printed them, against the indication the report attached them to, with this note beside them.
CorrectionColorectal cancer carries “Yes (0.49)” in that same column of the scorecard.de-crc-0.49
The differential-expression table has no colorectal row of any kind. The value 0.493 is gastric carcinoma’s, and the scorecard prints it again one row down against gastric. The ranking report’s tier table gives colorectal cancer 0.405.
How this site handles it
The value is reproduced as printed, with the two values the pack prints for colorectal cancer set beside it.
No table in either report prints 0.64 for any disease. It would be the largest fold change in the pack, above the 0.611 the ranking report gives basal-like breast carcinoma. The ranking report’s tier table prints “N/A” for glioblastoma’s fold change.
How this site handles it
The value is reproduced as printed and marked as unsupported.
CorrectionProstate is marked “Yes (0.31)” for differential expression.de-prostate-0.31
No table in either report prints 0.31 for any disease, and neither report’s expression table contains a prostate row.
How this site handles it
The value is reproduced as printed and marked as unsupported.
CorrectionClock count given as four in three places and as five in the table printed between two of them.clock-count-against-table
Where it appears
The ageing review, 3.1 Summary Table. Read against the executive summary, the opening of section 3 and the conclusion at 3.5.
Why it cannot stand
The count of 4 is printed in the executive summary, again in the sentence that opens the clocks section — “4 distinct aging clocks” — and again in the conclusion: “KIF18A appears in 4 aging clocks across methylation and transcriptomic modalities.” The table between the second and the third carries 5 rows: Mammalian life-history clock, AltumAge methylation clock, PASTA transcriptomic age-shift clock, REG transcriptomic age clock and ZhangBLUP methylation clock. The review never says which four of the five it means, so the difference cannot be settled from inside the document.
How this site handles it
Both figures are reproduced. The clocks section of this page draws every row the table prints, and says beside them that the review counts fewer.
CorrectionThree incompatible counts of one CpG set: eleven named in the summary table, twelve tallied in the cell beside them and fourteen tabulated one section down.zhangblup-cpg-counts
Where it appears
The ageing review, 3.3 ZhangBLUP Clock — Detailed CpG Analysis. Read against the ZhangBLUP row of the summary table at 3.1.
Why it cannot stand
The summary table’s feature cell reads “11 CpG sites” and its interpretation cell reads “(8 negative, 4 positive)”, which is 12. The detailed section opens by saying the clock contains “11 KIF18A-linked CpGs”, lists 14 of them with a coefficient and a sign each, and closes: “10 out of 14 CpG sites have negative coefficients”. Counting the signs in that table gives 10 negative and 4 positive. No two of the three counts agree, and the conclusion of the clocks section carries the 10-of-14 form forward without noting the others.
How this site handles it
The page prints the counts the review printed, in the places it printed them, and chooses between none of them.
CorrectionSeventeen significant tissues counted in the sentence beneath the expression table, one rows printed in the table above it, and the extra rising tissue is never named.gtex-tissue-counts
Where it appears
The ageing review, 4.1 Summary of Significant Correlations. The summary sentence read against the table it sits under.
Why it cannot stand
The sentence reads that out of 49 tissues tested, “17 showed significant correlations” at p below 0.05: “10 with negative (decreasing with age) and 7 with positive (increasing with age) trends”. The table above it prints 1 rows, 0 falling with age and 0 rising. The count of tissues falling matches the table exactly, so the whole of the difference sits in the rising column, and nothing in the section says which tissue the extra row would be.
How this site handles it
The chart on this page draws the 1 rows the table prints, because each of those carries a tissue, a correlation, a p-value and a sample size. The sentence’s own counts are printed beside the chart.
DiscrepancyATX020 attributed to two developers and three stages of development across four documents, with the first of them contradicted by its own reference list.atx020-developer
Where it appears
The thirteen-indication ranking, Competitive Landscape, Preclinical Tools and Academic Compounds. Read against section 2 of the research directive and page 5 of the clinical evidence review.
Why the two differ
The ranking’s table of preclinical tools gives ATX020 the developer “Athenex / NCI” and the status “Preclinical”, and cites two 2025 papers for it. The research directive asks a later reader to “correct the ATX020 attribution”, on the ground that those same two papers, at PMIDs 41257005 and 41369352, are “both Accent Therapeutics work”. The clinical evidence review puts an “AMG 650 / ATX020 series” in its precedent table at a different stage again, “Early clinical”. Insilico’s indication prioritisation report prints the code a fourth way, as a bracketed alternative name for a Phase 1 asset of Accent Therapeutics.
How this site handles it
The competitive account on this page follows the two clinical reports, which resolve each programme to a registry identifier. The ranking’s row is reproduced with the developer it names, and the directive’s objection is printed beside it rather than applied to it.
Note on the sourceProstate appears twice in the same fifteen-row list, as “prostate adenocarcinoma” at rank 33 and “prostate carcinoma” at rank 26, both carrying a total score of 5.8 and the same first two drivers, PPI (0.96) and matrix factorization (0.90).prostate-near-duplicate
Only the third driver differs, GWAS sub-modules at 0.75 against 0.89. The report’s own methodology says umbrella and ontology-parent terms are excluded from ranking, tiers and top lists.
How this site handles it
Both rows are printed as the report printed them, side by side.
Note on the sourceThe indication report puts rank 19 on papillary renal cell carcinoma. The ranking report’s tier 3 table puts rank 19 on clear cell renal cell carcinoma.rank-19-rcc-subtype
Where it appears
The thirteen-indication ranking, Tiered Indication Ranking, Tier 3. Rank 19, against the ranking report’s tier 3 table.
Why it needs care
The two reports name different kidney cancer subtypes at the same PandaOmics rank, and the fold change of 0.209 in the ranking report is labelled clear cell renal carcinoma in its own expression table.
How this site handles it
Both subtypes are carried as separate rows, each with the rank its own report gives it.
Note on the sourceThe tier 2 table prints “N/A” for gastric cancer’s PandaOmics rank.gastric-rank-na
The ranking report prints the same disease and the same value, -0.099 at p = 0.005, as “Hodgkin’s lymphoma”.
How this site handles it
Each report’s spelling is reproduced where that report is cited.
Note on the sourceThe dimension table scores Expression (omics) at 0.156 and reads it “LOW”. The dataset table on the same page reads the combined expression result as “No age-associated KIF18A expression signal” with a significance of “Absent”.expression-omics-0.156
Where it appears
The clinical translation assessment, page 5.Section 4.2, the dimension score table and the dataset table above it.
Why it needs care
A score of 0.156 is a low signal and “Absent” is no signal. The page states both about the same measurement without reconciling them.
How this site handles it
Both readings are printed, from the same page, as the report gives them.
Note on the sourcePASTA and REG carry opposite coefficient signs on the identical gene identifier ENSG00000121621 from one 2025 preprint, and the summary table counts both in favour of inhibition.transcriptomic-clock-signs
Where it appears
The ageing review, 3.4 Transcriptomic Clocks (PASTA and REG). The two rows it explains, printed side by side in the summary table at 3.1.
Why it needs care
Both rows print the identifier ENSG00000121621, and the section beneath the table attributes both clocks to the same authors and the same 2025 preprint. The first carries a negative coefficient and is tagged “SUPPORTS inhibition”, on the reading that higher expression goes with a younger transcriptomic age. The second carries a positive coefficient and reads “Inhibition would reduce this contribution.” The section says the second “suggests the opposite” of the first, and neither row is withdrawn. Two clocks disagreeing about one gene is an ordinary result; what needs care is that the table books both as support.
How this site handles it
Both rows are drawn with the sign each carries, and the page says which of the two the review tagged and which it did not.
Note on the sourcecg14927277 counted twice, once as the whole of the AltumAge feature cell and once as a site in the ZhangBLUP table, with the two clocks then counted as separate lines of evidence.shared-cpg-across-clocks
Where it appears
The ageing review, 3.3 ZhangBLUP Clock — Detailed CpG Analysis. Read against the AltumAge feature cell in the summary table at 3.1.
Why it needs care
The site cg14927277 is the entire feature cell of the AltumAge row, and it appears again in the detailed table one section down, where it carries a positive coefficient. The summary that closes the section counts the two clocks as separate findings. Nothing is wrong with one site appearing in two models. What a reader needs is to know that it does before treating the clocks as independent measurements.
How this site handles it
The overlap is stated where the clocks are drawn. Neither row is removed and neither clock is recounted.
Note on the sourceThree partial responses plus six with stable disease is nine, against seven of seventeen evaluable patients credited with tumour reduction, all in one sentence.vls1488-readout-counts
Where it appears
The clinical evidence review, page 3. The single bullet reporting the 2025 conference abstract for VLS-1488.
Why it needs care
The bullet reads: “7/17 (41%) evaluable HGSOC patients with tumor reduction; 3 partial responses; 6 with stable disease”. The readout table on this page records the arithmetic as: 3 partial responses plus 6 with stable disease is 9 patients, which is more than the 7 credited with tumour reduction out of 17 evaluable. The three counts cannot describe one set nested inside another. They need not: a patient whose disease is stable may have had no measurable shrinkage. The sentence does not say which reading applies, and the same page records that these figures arrived as an uploaded dossier and were never independently checked.
How this site handles it
All three counts are printed as the review printed them, in one table, with the arithmetic set out beside them. Nothing on this page is computed from them.
Note on the sourceAgeing indications ranked between 432 and 856 in a superseded report and at 40 and 37 in the one this pack holds, with only the second of the two reports available to check.ageing-rank-across-versions
Where it appears
The research directive, Core Objective, 1. Source Adjudication (complete before building on any report). The paragraph headed “Aging rank discrepancy across versions.”.
Why it needs care
The directive writes that a superseded earlier version “placed the classical aging indications far down the 1,000-item list”, and names Alzheimer’s disease at 432, skin ageing at 513, Parkinson’s disease at 546, osteoporosis at 619 and age-related macular degeneration at 856. It sets those against the version in this pack, which it reports as neurodegenerative disease at 40 and diabetes mellitus at 37, both flagged No-go. The second pair is checkable here: the verdict table of Insilico’s indication prioritisation report prints both ranks and both calls. The first five are not, because the superseded report is not one of the six documents in this pack. The directive asks whoever reads it to “Establish whether this is an ontology-term difference or a scoring change, and state which ranking you are relying on.” No document in the pack answers that.
How this site handles it
This page relies on the version it holds and says so. The earlier ranks are named here, where their provenance is stated, and are drawn nowhere else.
Why these are listed
None of these is a reason to discard the document it belongs to. They are the reasons to read its headline numbers against its tables before quoting them.
How this page was built
16. Method, Scope and Provenance
Six internal documents, no outside search, and every figure traced to its source before the build completes.
This page was assembled from six internal documents and nothing else. No literature search was run, no database was queried, and no number appears here that is not either in one of those documents or worked out from them in the open.
Where the documents agree, the page says so once. Where they disagree, both positions are printed with the place each was read, and the disagreement is left standing. Nothing was reconciled, averaged, or quietly dropped.
Rules this page follows
Every figure on the page is traced back to the document it came from before the build completes; a figure that cannot be found in a cited source stops the build.
A number worked out here rather than read is declared as such in the build file, with the working written beside it.
Quotations are checked character by character against the extracted text of the document they are attributed to.
Defects found in the sources are recorded and never corrected, because a repaired number would leave the reader unable to check the page against the report it came from.
Content in the sources that is not on this page
Anything not in the six documents, including the primary literature they cite. Where a report describes a paper, this page reports what the report said about it, not what the paper says.
The conference abstract behind the one efficacy readout in the pack, which the clinical evidence review received as an upload and states it did not verify against a primary source.
What this page cannot tell you
Five of the six documents were produced by the same family of internal tools, so agreement between them is weaker evidence than it looks.
Two documents report positions out of differently sized fields of indications, and neither reconciled its numbers with the other’s. Both are carried.
One document’s page locators are section numbers rather than printed pages, so its citations point to sections.
Sources cited
17. References and Terms
Terms used here
Aneuploidy
Having the wrong number of chromosomes in a cell. A healthy human cell carries forty-six; a cell that has gained or lost one is aneuploid. It accumulates with age and is close to universal in the tumours these programmes target.
Chromosomal instability
A cell’s tendency to keep gaining and losing chromosomes each time it divides. It is the process; aneuploidy is the result. The trials in this pack select patients on it. Written in the reports as: CIN, CIN-high, CIN-low
Kinesin
A family of motor proteins that walk along the internal fibres of a cell, carrying cargo or adjusting the fibres themselves. KIF18A is one of them, and the part that does the walking is called the motor domain.
Microtubule
The stiff hollow fibres a cell builds to pull its chromosomes apart when it divides. KIF18A walks along these and controls how fast their ends grow and shrink.
Synthetic lethality
When losing either of two things is survivable but losing both is not. Here it is the argument for the whole target: normal cells tolerate the loss of KIF18A, and cells already struggling with unstable chromosomes do not.
Senescence
A state in which a damaged cell stops dividing but does not die, and instead sits in the tissue releasing inflammatory signals. A senolytic is a drug that kills such cells. Written in the reports as: senescent cells, senolytic
Senescence-associated secretory phenotype
The mixture of inflammatory proteins a senescent cell releases, which is how one worn-out cell makes the tissue around it worse. Written in the reports as: SASP
Epigenetic clock
A statistical model that estimates a person’s age from chemical marks on their DNA. Several of the reports in this pack use the weight a clock gives KIF18A as evidence that the gene matters to ageing.
Log2 fold change
How much more, or less, a gene is switched on in diseased tissue than in healthy tissue, on a doubling scale. A value of one means twice as much; a negative value means less. Written in the reports as: logFC
Geroscience endpoint
A measurement in a clinical trial that is about ageing itself rather than about one disease, such as a biological-age estimate or a count of senescent cells. No trial in this pack carries one.
Patient-derived xenograft
A piece of a patient’s tumour grown in a mouse, used to test whether a drug works against that particular tumour before trying it in people. Written in the reports as: PDX
Companion diagnostic
A test used alongside a drug to decide who should receive it. For this target it would be a measurement of how unstable a tumour’s chromosomes are, and no threshold for it has been fixed.