drug-repurposing

name	drug-repurposing
description	Systematic drug repurposing via signature matching, target-based analysis, network proximity, genetic evidence scoring, and clinical evidence mining
version	1.0.0
tags	["drug-repurposing","repositioning","cmap","connectivity-score","network-proximity","genetic-evidence","indication-expansion"]

Drug Repurposing

This skill guides systematic identification of existing drugs for new therapeutic indications using computational methods. Covers signature- based, target-based, network-based, and genetics-based approaches.

Strategy Selection Decision Tree

Choose the repurposing strategy based on available data:

1. Signature-based repurposing (have: disease DE signature)

   • Query a disease transcriptomic signature against drug perturbation profiles (CMap-style connectivity scoring).
   • Drugs that REVERSE the disease signature are therapeutic candidates. Drugs that MIMIC it may exacerbate.
   • See references/repurposing-methods.md for connectivity scoring with gseapy.prerank and permutation testing.
   • Input: ranked gene list from DE analysis (sign(log2FC) * -log10(pvalue)), plus a set of drug perturbation profiles.
   • Reference-data caveat: a drug-perturbation signature collection is in the reference inventory as an opt-in download, so resolve it up front and expect it may not be staged. It arrives as directional gene sets — one up set and one down set per experiment — and connectivity is the difference between their two enrichment scores, so pair them by their shared term prefix and never score one alone. Failing that, search the workspace for staged perturbation data. If neither is present, say so and switch to a target-, network-, or genetics-based strategy — do not invent a signature path, and do not substitute drug-target gene sets for perturbation profiles and still call the output a connectivity score.
   • Output: ranked drugs by connectivity score with FDR.

2. Target-based repurposing (have: validated target gene list)

   • Map disease-relevant targets to known drugs via ChEMBL, DrugBank, and Open Targets.
   • Workflow: a. Resolve gene symbols to ChEMBL target identifiers. b. Find compounds with bioactivity against each of those targets. c. Check approval status and existing indications for each drug. d. Verify the mechanism of action is relevant to the disease. e. Where DrugBank lookup is available, add indication, interaction, and pharmacology data.
   • Prioritize: approved drugs > Phase 3 > Phase 2 > Phase 1 > preclinical.
   • Flag drugs already indicated for the target disease (not repurposing candidates).

3. Network-based repurposing (have: disease module / PPI network)

   • Compute network proximity between drug target sets and disease gene modules in a PPI network.
   • See references/repurposing-methods.md for network proximity scoring.
   • Input: disease gene set + drug-target mapping + PPI network.
   • Reference-data caveat: a genome-scale scored PPI network is in the reference inventory as an opt-in download, so resolve it before planning on it and expect it may not be staged. It is keyed on internal protein identifiers and ships a companion mapping table — read the entry's stated contents rather than assuming symbols. Score it to high confidence before measuring distances: the unfiltered graph is dense enough that proximity stops discriminating. If nothing resolves, and none is staged in the workspace either, report the gap rather than constructing a placeholder network — proximity z-scores from an invented graph are meaningless.
   • Output: drugs ranked by proximity z-score.
   • Closer proximity = more likely to modulate disease biology.

4. Genetics-based repurposing (have: GWAS hits or genetic associations)

   • Drugs targeting genes with genetic evidence for a disease have higher clinical success rates.
   • Workflow: a. Search the GWAS Catalog for genetic associations with the disease trait. b. Where DisGeNET lookup is available, add gene-disease association scores. c. Map the implicated genes to druggable targets via Open Targets, checking tractability. d. Find existing drugs for those targets in ChEMBL/DrugBank.
   • Prioritize targets with both genetic evidence AND existing drugs.

5. Clinical evidence mining (validation layer)

   • For any candidate, check real-world evidence: a. Search clinical-trial registries — is the drug already being tested for this indication? b. Query FAERS adverse-event reports — are there safety signals specific to the proposed patient population? c. Search the published literature — is there evidence for the drug-disease combination?
   • Report existing clinical evidence alongside computational predictions.

Multi-Strategy Integration

The strongest repurposing candidates are supported by multiple independent lines of evidence:

Evidence Type	Source	Weight
Signature reversal	CMap-style connectivity	Mechanistic
Genetic association	GWAS/DisGeNET	Causal inference
Target druggability	ChEMBL/Open Targets	Feasibility
Network proximity	PPI-based	Biological context
Clinical precedent	ClinicalTrials.gov	Translational
Safety profile	FAERS	Risk assessment
Literature support	PubMed	Prior knowledge

Scoring Framework

When combining evidence from multiple strategies:

1. Score each evidence type independently (normalize to 0-1).
2. Weight by evidence quality (genetic > signature > network for target validation; signature > network > genetic for mechanism).
3. Report composite score with component breakdown.
4. Flag candidates with >= 3 independent evidence lines as high-priority.

Compound Prioritization Criteria

After candidate identification, rank by:

Criterion	Priority	Source
Approved for another indication	Highest	DrugBank/ChEMBL
Active clinical trials (Phase 2+)	High	ClinicalTrials.gov
Known safety profile	High	FAERS, DrugBank
Mechanism matches disease biology	High	Literature, pathway analysis
Genetic evidence supports target	High	GWAS, DisGeNET
Favorable ADMET profile	Medium	ChEMBL ADMET, RDKit descriptors
No IP/exclusivity barriers	Medium	DrugBank status
Preclinical evidence only	Lower	ChEMBL bioactivity

Reporting Standards

Every repurposing analysis must report:

1. Strategy used — which approach(es) and why.
2. Disease signature or gene set — exact definition, source, size, and how it was derived.
3. Reference database versions — CMap build, ChEMBL version, GWAS Catalog date.
4. Statistical framework — permutation testing for connectivity scores, multiple testing correction for multi-target analysis.
5. Candidate table — drug name, existing indication, proposed indication, evidence score, evidence sources, development stage.
6. Validation evidence — clinical trials, literature, safety data.
7. Limitations — computational predictions require experimental validation; repurposing candidates are hypotheses, not validated therapeutics.

References

Reference	File	Contents
Repurposing Methods	references/repurposing-methods.md	CMap connectivity scoring with gseapy, network proximity algorithm, genetics-based target scoring, multi-evidence integration, candidate ranking

Do NOT

• Present computational repurposing candidates as validated therapeutics — they are hypotheses requiring experimental and clinical validation
• Use connectivity scores without permutation-based significance testing — raw enrichment scores are not interpretable without a null distribution
• Ignore existing indications — a drug already approved for the target disease is not a repurposing candidate
• Rank candidates by a single evidence type — multi-evidence integration is essential for meaningful prioritization
• Skip safety assessment — even approved drugs may have contraindications for the proposed patient population
• Claim "drug X treats disease Y" — say "drug X reverses the disease Y transcriptomic signature" or "drug X targets a gene genetically associated with disease Y"
• Use outdated perturbation databases without noting the version and its limitations
• Ignore drug-drug interactions when proposing repurposed drugs for patients likely on existing therapies