drug-repurposing

Overview

drug-repurposing covers the systematic search for new uses of existing drugs: computational prediction (signature matching, network approaches, structure-based), evidence triangulation across data types, and the validation path from prediction to clinical testing. Repurposing is attractive (known safety, faster trials) but littered with false positives — the skill is about separating plausible candidates from computational noise.

When to use

• Generating repurposing hypotheses for a disease or target.
• Signature-based screening: disease signatures reversed by drug signatures (Connectivity Map logic).
• Network-based repurposing: drug-target-disease networks, proximity measures.
• Real-world evidence: EHR/pharmacoepidemiology signals for repurposed indications.
• Prioritizing candidates: safety, PK, IP, and trial-feasibility filters.
• Designing validation: from in vitro to observational to interventional.

Core concepts

• Why repurposing works (sometimes). Known human PK/safety de-risks development; the biology rationale is that drugs hit pathways shared across diseases. It works best when the mechanism is specific and the new indication shares the pathway — worst when it's pure computational association.
• Signature reversal. Disease gene-expression signature vs drug-induced signatures (Connectivity Map/LINCS): drugs that invert the disease signature are candidates. Caveats: cell-line signatures ≠ human disease; reversal correlation is weak evidence alone; validate the signature's disease relevance first.
• Network proximity. In drug-target-disease networks, efficacious drugs tend to target proteins proximal to disease modules. Useful for hypothesis generation; the networks are incomplete and biased toward well-studied proteins (which conveniently are also the druggable ones).
• Real-world evidence. EHR analyses asking "do users of drug X have better outcomes for disease Y?" — subject to massive confounding by indication. Use active comparators, new-user designs, and negative controls (see ehr-data-analysis). RWE generates hypotheses; it doesn't confirm them.
• Triangulation. A candidate supported by signature reversal AND network proximity AND genetic evidence (Mendelian randomization implicating the target) AND RWE is worth pursuing. Single-evidence candidates are usually noise — require convergence.
• Practical filters. Safety at the needed dose/duration (a cancer drug's safety profile may not transfer to chronic use); PK (does it reach the target tissue? CNS penetration for neuro indications?); formulation and IP status; and whether anyone will fund the trial.
• Mendelian randomization as target validation. Genetic variants mimicking drug action (e.g. in the target gene) associated with disease risk support the target. One of the strongest human-genetics validations available — check MR evidence for top candidates.
• Validation ladder. In vitro disease-relevant assay → in vivo model (with PK confirming exposure) → observational/human-genetics support → small proof-of-concept trial. Skipping rungs is how repurposing fails expensively.

Practical workflow

1. Define the target profile. Disease, desired mechanism, acceptable safety, target tissue, and commercial/practical constraints — before screening.
2. Computational screen. Signature reversal and/or network proximity across approved drugs; rank transparently with all scores reported.
3. Triangulate. MR evidence, RWE signals, literature — require ≥2 independent evidence lines for shortlisting.
4. Filter. Safety at indication-appropriate exposure, tissue PK, IP/regulatory path, trial feasibility.
5. Validate experimentally. Disease-relevant assays (not just easy cell lines); confirm target engagement and exposure in vivo.
6. Clinical path. Proof-of-concept trial design with biomarkers of target engagement; consider platform/basket designs for efficiency.
7. Report. Full ranked list (not just hits), methods, negative results — the field needs the misses too.

Common pitfalls

• Single-evidence candidates (pure computational hits) advanced as discoveries.
• Cell-line signatures treated as disease truth.
• Confounding by indication in RWE repurposing signals.
• Safety profiles assumed to transfer across doses, durations, and populations.
• CNS indications for drugs that don't cross the blood-brain barrier.
• No IP/regulatory path — great science nobody can develop.
• Publication bias: only positive repurposing stories get told.