Are drug targets with genetic support twice as likely to be approved? Revised estimates of the impact of genetic support for drug mechanisms on the probability of drug approval.

Key results

1. ShownMendelian (OMIM) genetic support replicates on data unseen by Nelson et al., with effects matching or exceeding the original estimates except for Phase I to II progression, and it predicts clinical advances that occurred after the 2013 data freeze.

   Pipeline Progression set (5513 pairs testable since 2013): OMIM support significantly raised Phase II to III progression; OMIM estimates comparable to or above Nelson et al. except Phase I to II; effect stays significant after excluding targets with pre-2013 approvals for similar indications (Tables S11-12); Full Data OMIM effect exceeds original. · 3.1 Identifying Validation Sets; 3.2 Estimated Effect of Genetic Evidence on…; 4 Discussion

2. ShownGWAS Catalog support is a weaker predictor: in independent validation sets its effects are mostly indistinguishable from null, though across the full updated dataset it still significantly raises the probability of reaching approval from Phase I.

   All validation-set GWAS risk-ratio 95% CIs overlap one, except a negative effect on Phase II to III in Pipeline Progression; New Genetic estimates often significantly below Nelson et al.; Full Data GWAS effect on Phase I to Approval significantly positive; estimate highly sensitive to similarity cutoff (Figure S3). · 3.2 Estimated Effect of Genetic Evidence on…; 3.3 Statistical modeling of genetic effect on…; 4 Discussion

3. ShownTreating genetic evidence as a continuous trait-indication similarity and adjusting for confounders, approval odds rise with similarity for both sources, while targets genetically linked only to dissimilar traits fare worse than targets with no genetic link.

   Bayesian logistic regression in Stan on 20292 pairs with US/EU approved or inactive drugs, adjusting for RVIS, development time, target class/GO terms and top-level MeSH category; p = 2 supported by WAIC; 95% credible intervals for GWAS and OMIM similarity effects exclude zero. · 3.3 Statistical modeling of genetic effect on…; 5.5.2 Bayesian Logistic Regression; Figure legends

4. SuggestedWhen the causal gene is clear, GWAS evidence behaves much like OMIM: restricting to lead SNPs in tight LD with coding or predicted deleterious variants moves the GWAS effect toward the OMIM level, pointing to SNP-to-gene assignment as a bottleneck.

   Restriction to GWAS Catalog SNPs with R^2 ≥ 0.9 to missense or moderately/highly deleterious variants; abstract reports approval from Phase I more than two-fold higher with clear causal genes; stricter LD cutoff gives larger missense effect (Figure S24); no numeric estimate in main text. · Abstract; 3.3 Statistical modeling of genetic effect on…; 4 Discussion

5. ShownRe-deriving historical development phases from database event fields recovers pipeline records that Nelson et al. discarded as inactive, greatly expanding the set of analyzable target-indication pairs.

   83% of inactive-status records became usable; 26360 pairs mapped to a highest phase vs 8853 labelled in Nelson et al.; 14759 pairs absent or inactive in 2013 newly phased; standardized MeSH mapping raised overlap with Nelson pairs from 88% to 98%. · 3.1 Identifying Validation Sets; 5.1 Pipeline data; 5.3 Trait-indication similarities