Results
methods | index | discussion
3.1 Overall
Of 14 claims: 8 supported, 3 partly supported, 2 not testable in the app, and 1 missed by the co-pilot.
Sources: manuscript Supplemental Data §S5; Supporting_Data/tables/claims_vs_app_evidence.csv.
3.2 KRAS is the right backbone (slide 39)
| Metric | Value |
|---|---|
| Rank | #2 of 6,000 targets |
| Overall score | 100 |
| Verdict | Top-tier candidate |
| DepMap Chronos | −2.14 |
| Pancreatic lines dependent | 96% |
| Mutated in tumours (cBioPortal) | 65.4% |
| Tractability | 100 |
KRAS is the strongest dependency of the seven genes.
Evidence: KRAS_header.png · KRAS_dependency.png · full KRAS target report (supporting-data)
3.3 FAK is a standing dependency (slides 38 and 40)
| Metric | App (snapshot #143 v9) | Case study | Agreement |
|---|---|---|---|
| DepMap Chronos | −0.64 | −0.616 | within 0.03 |
| Pancreatic lines dependent | 58% | 57% | within 1 pp |
| Dependency score | 64 / 100 | — | — |
FAK is the second strongest dependency of the seven genes, after KRAS.
FAK’s overall rank is only #293, because it scores 0 on genetics (mutated in 0.6% of tumours) and 21 on expression. That is the case study’s point on slide 40: FAK does not look prominent, but tumours need it.
Evidence: PTK2_header.png · PTK2_dependency.png · full PTK2 target report (supporting-data)
3.4 Network prominence does not predict dependency (slide 40)
All seven genes score between 94 and 100 on the app’s network measure (WINNER centrality percentile), so the network cannot tell them apart. On dependency they range from 0 (FN1) to 100 (KRAS). KRAS and FAK are the only two genes with a dependency score of 50 or more.
Evidence: network_vs_dependency.png · criterion_scores_heatmap.png (supporting-data)
3.5 SRC organises the network but is not a dependency (slides 35 and 36)
| Metric | Value |
|---|---|
| Network centrality | 99.8th percentile (Network 100) |
| Interaction partners | 964 |
| DepMap Chronos | −0.22 |
| Pancreatic lines dependent | 15% (Dependency 22) |
The SRC neighbours named on slide 35 — FN1, EGFR, ERBB3, FAK — all score 94 to 100 on network centrality.
Evidence: SRC_network.png · SRC_dependency.png · full SRC target report (supporting-data)
3.6 The ERBB escape route is druggable (slide 37)
| Gene | Rank | Tractability | Selective drugs | ChEMBL compounds | Pancreatic trials (Phase 2 precedent) |
|---|---|---|---|---|---|
| ERBB2 | #72 | 100 | 39 | 48 | 36 |
| ERBB3 | #108 | 100 | 15 | 17 | 13 |
The induction of ERBB2/3 after KRAS blockade is not something the app can show; it comes from the lab’s perturbation datasets.
Evidence: ERBB2_tractability.png · ERBB3_tractability.png · full target reports: ERBB2, ERBB3 (supporting-data)
3.7 Clinical precedent (slide 33)
| Gene | Trials in pancreatic cancer | Max phase | Stopped | Stopped for safety/efficacy |
|---|---|---|---|---|
| SRC | 15 | Phase 2 | 3 | 1 |
| FAK (PTK2) | 14 | Phase 2 | — | none |
The SRC picture matches the case study’s statement that SRC-directed trials have had uncertain outcomes.
Evidence: SRC_clinical.png · PTK2_clinical.png (supporting-data)
3.8 Co-pilot (slide 67)
In both runs the co-pilot:
- listed both Phase 3 RASolute trials with correct IDs —
NCT06625320andNCT07491445. Slide 67 had recorded a wrong ID for the second trial, so this has improved; - found no registered trial testing daraxonrasib with defactinib;
- listed five related defactinib trials and five publications with PMIDs.
In both runs it missed AACR 2026 Abstract 6497 on KRAS + FAK synergy (Cancer Res 2026;86(7_Suppl):6497), which slide 67 cites. Conference abstracts may not be indexed in Europe PMC.
3.9 Score reproduction
All 49 criterion scores (7 genes × 7 criteria with full inputs) and all 7 overall scores were reproduced from their raw inputs using the app’s formulas, within one point of rounding.
Source: manuscript Supplemental Data §S4.
Comments