Genetic evidence improves chances of drug discovery success. M. R. Nelson1, H. Tipney2, J. L. Painter1, J. Shen1, P. Nicoletti3, Y. Shen3,4, A. Floratos3,4, P. C. Sham5, M. J. Li6, J. Wang6, P. Agarwal7, J. C. Whittaker2, P. Sanseau2 1) Quantitative Sciences, GlaxoSmithKline, Research Triangle Park, NC, USA; 2) Quantitative Sciences, GlaxoSmithKline, Stevenage, UK; 3) Initiative in Systems Biology, Columbia University, New York, NY, USA; 4) Department of Biomedical Informatics, Columbia University, New York, NY, USA; 5) Centre for Genomics Sciences, University of Hong Kong, Hong Kong SAR, China; 6) Department of Biochemistry, University of Hong Kong, Hong Kong SAR, China; 7) Quantitative Sciences, GlaxoSmithKline, Upper Merion, PA, USA.

   Attrition is a major challenge in drug discovery and development with >90% of projects failing before clinical trials and >50% of the remainder failing in clinical development due to lack of efficacy. Therefore, selecting and validating the best targets is the key decision in developing medicines. Often the human evidence supporting the chosen target in a disease context is limited. However, rapid progress in deciphering the genetic basis of disease, and associated pathways, offers an opportunity to transform this process and leads to a key question: what weight should be given to a genetic association in selecting targets and indications? To address this question we merged GWASdb, a database with over 100000 genetic associations corresponding to 1228 unique traits mapped to 603 Medical Subject Heading (MeSH) terms, with Informa Pipeline, a commercial database of >23000 drugs with known human targets (including >2400 marketed) with 915 unique indications mapped to 708 MeSH terms. We drew on linkage disequilibrium, eQTL data, ENCODE-related data, and location to map genetic variants to one or more genes. We mapped 14614 genome-wide significant variants (p < 1e-8) to 3781 protein-coding genes (19%). We found that among 1855 unique drug targets in Pipeline, 522 (28%) had one or more genome-wide significant associations. Interestingly, targets for marketed drugs had a significantly higher percentage of genes connected to genome-wide significant associations (128 of 379 genes, 34%). Furthermore, targets for marketed drugs were 8 times more likely to be in OMIM or have a genetic association (N = 221; 58%) compared to all coding genes in the genome, demonstrating that genetic evidence at the target increases the chance that a drug will reach the market. We used the MeSH hierarchy to estimate relative similarities between terms to investigate the similarity between drug indication and genetically associated trait: as term similarity increases, so does the proportion of marketed drugs. Conversely, the many marketed drugs that have different indications than the genetically associated traits may suggest exciting opportunities for new indications. This work also identifies many drug indications for which there are no genetic associations, highlighting opportunities for future genetic studies. We conclude that genetic associations should play an important role in making decisions about target selection and indications to be investigated in drug development.

You may contact the first author (during and after the meeting) at