TruthNexus
Drug Discovery · Medical

Every candidate explains why it ranked.

90% of drug candidates fail in clinical trials. Target identification still relies on manual literature review — a process that cannot systematically integrate genomics, proteomics, and clinical outcomes simultaneously. Skippy combines 11 peer-reviewed scoring methods across 74.7M verified findings, with 10 explainability overlays that show the evidence path from drug to mechanism to disease.

GET /api/proximity · 200 OK
Metformin
Alzheimer's Disease
0.78
composite
14
ABC paths
7/11
concordant
0.65
novelty
Key intermediaries
AMPKmTORPI3KFOXO3SIRT1
OMIM ×2GWAS 1 locusL1000 corr −0.42
74.7M
Fused findings across 50+ sources
11
Peer-reviewed scoring methods synthesized
10
Explainability overlays per candidate
90%
Of drug candidates fail — this is why
See It In Action

What a response looks like

Representative API output across three core use cases — repurposing, deep pair analysis, and combination scoring.

POST /api/repurpose · application/jsonsimulation
Alzheimer's Disease
11 scoring methods · temporal holdout benchmark
Candidate
Score
ABC paths
Novelty
1Metformin7/11
0.78
140.65

AMPK activation → mTOR suppression → autophagy enhancement; neuroprotective in multiple animal models

2Rapamycin6/11
0.71
110.72

mTOR inhibition → reduced tau hyperphosphorylation and amyloid-β production

3Riluzole5/11
0.65
80.81

Glutamate release inhibition → reduced excitotoxicity in hippocampal neurons

4Pioglitazone5/11
0.61
70.69

PPARγ agonism → reduced neuroinflammation via microglia modulation

5Lithium4/11
0.58
90.55

GSK-3β inhibition → reduced tau phosphorylation; established neuroprotective evidence

Simulated output representative of real API responses. 11 scoring methods, 50+ sources, temporal holdout benchmark. Research use only.

What It Does

Six capabilities. One API.

POST /api/repurpose

Drug repurposing

Enter a disease; receive ranked drug candidates combined across 11 scoring methods — Swanson ABC path score, DWPC hub-corrected, genetic evidence, signature reversion, RWR PageRank, network proximity, and more.

GET /api/proximity

Pair analysis

Deep analysis of a specific drug–disease pair: ABC path count and p-value, network proximity, genetic evidence (OMIM, GWAS, DisGeNET), L1000 signature reversion correlation, novelty score, and key intermediary genes.

GET /api/drug-similarity

Combination scoring

Score drug pairs by target complementarity and Jaccard overlap. Low Jaccard over shared intermediary genes = orthogonal mechanisms = combination potential. Cross-validated against NCI ALMANAC synergy data.

GET /api/genetic-evidence

Genetic evidence

GWAS Catalog, OMIM, DisGeNET, and ClinVar evidence for a gene–disease association, returned as structured confidence scores. Used as a scoring component and available as a standalone lookup.

GET /api/pathways

ABC pathway explorer

Enumerate Swanson ABC paths between a drug and disease through shared intermediary genes. Returns path count, top intermediaries, and hypergeometric p-value for shared-gene significance.

POST /api/batch/run

Batch pipeline

Run the full discovery pipeline across a list of diseases in parallel. Export results as structured JSON or CSV — useful for pipeline screens, indication expansion, and systematic portfolio analysis.

Scoring Methods

11 scoring methods. One composite.

Composite = weighted sum of 11 peer-reviewed methods × (1 − ABC p-value). Weights are configurable via the Hypothesis Workbench (M9). Temporal holdout benchmark — not random splits, which inflate AUC by 5–15%.

ABC path score18%
DWPC hub-corrected14%
Genetic evidence (OMIM / ClinVar / DisGeNET)14%
Signature reversion (L1000 / LINCS)10%
Related disease boost10%
Edge confidence weighting10%
RWR PageRank8%
Multi-scale diversity (Gysi 2021)8%
Path diversity8%
Temporal holdout benchmark

Validated on 35 curated known drug-disease pairs using temporal holdout — training on evidence before a cutoff date, validating on pairs discovered after. Random splits inflate AUC by 5–15%; temporal holdout does not. Recall@10, Recall@50, MRR, with 95% bootstrap confidence intervals.

0.71
Recall@10 temporal holdout
Validation

Classic discoveries that anchor the benchmark

Swanson's 1986 fish oil → Raynaud's discovery is where ABC-path repurposing began. These known pairs validate the methodology — if the engine can't rediscover them, it shouldn't be trusted on unknowns.

Fish Oil
Raynaud's Disease
Validated

Swanson's original discovery: omega-3 fatty acids connected via blood viscosity pathways.

1986
Metformin
Alzheimer's Disease
In trials

AMPK activation → mTOR suppression → autophagy enhancement. Multiple ongoing clinical trials.

Ongoing
Baricitinib
COVID-19
FDA EUA

JAK inhibition attenuates cytokine storm while preserving antiviral response. FDA EUA granted.

2021
Rapamycin
Aging / Longevity
In trials

mTOR inhibition links to multiple age-related diseases via autophagy and cellular senescence.

Ongoing
The benchmark problem

Most published drug repurposing systems report AUC on random train/test splits. Random splits allow drug-disease pairs from the same drug to appear in both train and test — inflating AUC by 5–15% without improving real-world discovery performance.

Skippy Drug Discovery uses temporal holdout: train on evidence before a cutoff date, validate on pairs first described after. Harder. Less impressive on paper. More informative about real performance.

The explainability gap
“A ranked list without mechanism is a ranked list. Every Skippy candidate shows the intermediary genes, the methods that agree, the methods that disagree, and the evidence path from drug to disease.”

Skippy Drug Discovery · 10 explainability overlays per candidate

Explainability

10 overlays. Every candidate.

Not post-hoc explanations bolted on after scoring — each overlay is a first-class output of the discovery pipeline, computed alongside the score.

M1Prediction Registry

SHA-256 tamper-evident pre-registration log. Timestamp predictions before outcomes are known — rigorous prospective validation.

M2Method Disagreement

Flag candidates where scoring methods diverge. High disagreement = lower confidence; surface it rather than hide it.

M3Mechanism Coherence

Score the biological plausibility of the drug-disease connection based on pathway coherence across the evidence base.

M4Counterfactual

"What if this gene were not shared?" — identifies which intermediary nodes are load-bearing for the repurposing hypothesis.

M5Failure-Mode Diagnosis

Explains why a candidate ranks low: poor proximity, missing genetic evidence, or no signature reversion signal.

M7GWAS Ancestry Equity

Flags candidates whose genetic evidence is derived predominantly from a single ancestry cohort — a known source of generalizability bias.

M8Novelty Score

PubMed co-occurrence analysis: candidates with strong mechanistic signal but low prior publication overlap surface as high-novelty opportunities.

M9Hypothesis Workbench

9 weight sliders let you co-author the scoring model live. Results update instantly — no server call — as you adjust method weights.

M10Combination Potential

Jaccard-based target complementarity score for drug pairs. Low overlap over intermediary genes = orthogonal mechanisms = synergy candidate.

Who It's For

Pharma, biotech, and research teams

Biotech Companies
Systematically screen repurposing opportunities across your therapeutic area. 11 methods, ranked and explained — not a literature dump.
Pharma R&D
Supplement internal programs with AI-identified candidates. Mechanism coherence scoring and disagreement flags prioritize candidates worth wet-lab follow-up.
Academic Research Centers
Generate evidence-backed hypotheses with novelty scoring that surfaces unexplored mechanistic territory — useful for grant applications and publication strategy.
Venture Capital & Foundations
Independent evidence synthesis for scientific due diligence on drug discovery assets. Benchmark results, prospective prediction track record, honest failure accounting.
Traceability

Discovery logged to tamper-evident registry

All discovery results are logged to a tamper-evident prediction registry with SHA-256 hash chain and versioned evidence base snapshots — enabling full traceability from discovery recommendation to any subsequent regulatory submission.

FAERS pharmacovigilance signals are integrated at the discovery stage to flag known safety concerns early. Every candidate surfaces with its evidence confidence and mechanism coherence score — not just a ranked list.

SHA-256 Tamper-Evident RegistryFAERS IntegrationVersioned Evidence SnapshotsResearch Use Only
Research Use

Skippy Drug Discovery is a research tool for scientific hypothesis generation. It does not replace wet-lab validation, clinical judgment, or regulatory review. All repurposing candidates require independent experimental validation before any clinical or commercial application. Predictions are logged to a tamper-evident registry with timestamps — not guarantees of therapeutic efficacy.

See Drug Discovery in your pipeline

We work with pharma R&D, biotech, and academic research centers. Let's talk about your repurposing or target identification problem.