Methods
2.1 Data source
All app evidence comes from Disease2Target (https://target.smartdrugdiscovery.org/).
| Parameter | Value |
|---|---|
| Dataset | pancreatic adenocarcinoma (MONDO_0006047) |
| Snapshot | #143, version 9 |
| Snapshot created | 15 September 2026 |
| Candidate genes | 6,000 |
| Evidence rows | 59,457 |
| Evidence axes | 13 |
| Modality weighting | small-molecule (used throughout) |
| Evidence captured | 26 September 2026 |
2.2 Gene selection
We checked the seven genes the case study’s argument depends on:
| Gene | Role in the case study | Slides |
|---|---|---|
| KRAS | The driver gene and the backbone of the proposed combination | 39 |
| PTK2 (FAK) | The proposed standing dependency | 38, 40 |
| SRC | The starting point of the analysis | 30–36 |
| ERBB2, ERBB3 | The proposed induced escape route | 37 |
| EGFR, FN1 | Named with FAK and ERBB3 as SRC’s network neighbours | 35 |
2.3 How each value was obtained
For each gene we:
- Opened the pancreatic adenocarcinoma dataset and waited until all eight criteria had finished loading. This matters because the ranking changes while data are still arriving.
- Searched the gene on the Ranking Board. This pins the gene and opens its side panel, giving its rank, overall score, verdict and eight criterion scores.
- Generated the gene’s full target report with the app’s Report button. The report lists every raw metric behind each score, its source, and whether it is a measured fact or a model prediction.
- Captured the full report as screenshots (
Supporting_Data/full_target_reports/) and copied every value into a table (Supporting_Data/tables/gene_evidence_snapshot143.csv). - Recorded the gene’s rank on the app’s Evidence tab as well, because that tab uses a different score (see discussion §4.3).
2.4 Co-pilot check
Slide 67 reports a session with the app’s co-pilot. We repeated the same question twice:
“Are there other trials or publications proposing daraxonrasib and defactinib for pancreatic cancer?”
- Model:
gpt-5.6-luna - Tools: live ClinicalTrials.gov and Europe PMC
We compared each answer with the five claims on slide 67. The full answer is in Supplemental_Data/copilot_answer_2026-09-26.txt.
2.5 Claim classification
We listed 14 claims from slides 30–40 and 67 and classed each one as:
| Class | Meaning |
|---|---|
| Supported | The app holds direct evidence for the claim |
| Partly supported | The app supports part of the claim (for example, druggability but not the mechanism) |
| Not testable in app | The claim rests on data the app does not carry |
| Missed | The co-pilot failed to retrieve evidence that exists |
2.6 Score reproduction
To check that the app’s scores are traceable, we recomputed every criterion score and every overall score from the raw values in the reports, using the formulas in the app’s source code:
rankingBoard.ts— https://github.com/aimed-lab/DiseaseToGene, commit6fcd205
Formulas and worked examples are in the manuscript’s Supplemental Data, sections S2 and S4.
Comments