Part of the National Institutes of Health Common Fund program, the Illuminating the Druggable Genome Knowledge Management Center (IDG KMC) is tasked with accurate data wrangling, processing and analytics for target discovery (druggablegenome.net). The IDG KMC distils more than 100 protein-centric and over 40 gene-centric resources for over 20,000 curated human proteins, which can be queried in Pharos (pharos.nih.gov). Complex data integration shows that ~3% of the proteome are drug targets associated with mechanism of action (“Tclin”), while 1 in 3 human proteins lacks functional information and disease relevance (“Tdark”; see Fig 1). We have collected and integrated over 20 biological datasets across the major knowledge domains including: genomic, proteomic, functional, phenotypic, and interactions (biochemical, protein-protein, TF-DNA, etc.) into target knowledge graphs. Here we report novel target-disease associations based on machine learning models (random forest with gradient boosting), for several diseases. For example, in Alzheimer's disease, we started with 53 genes associated with Alzheimer’s disease (positives) and 3,952 genes associated with other pathologies (negatives) from OMIM (www.omim.org). The test set had 23 genes associated with Alzheimer's (positives) and 200 random genes not associated with Alzheimer's (negatives) from DISEASES (diseases.jensenlab.org). Our model correctly predicted 20 of 23 positives, and 159 of 200 negatives. Our initial results show the potential of this approach to articulate knowledge via computational assertions, with focus on disease associations for target and drug repurposing

AI-driven target selection in drug discovery - exploring the undiscovered country

256th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.