A phenome-guided drug repositioning through a latent variable model

摘要: The phenome represents a distinct set of information in the human population. It has been explored particularly its relationship with genome to identify correlations for diseases. also drug repositioning efforts focusing on search space most similar candidate drugs. For comprehensive analysis phenome, we assumed that all phenotypes (indications and side effects) were inter-connected probabilistic distribution this characteristic may offer an opportunity new therapeutic indications given drug. Correspondingly, employed Latent Dirichlet Allocation (LDA), which introduces latent variables (topics) govern distribution. We developed our model Side Effect Resource (SIDER). first LDA optimized based recovery potential through perturbing drug-phenotype matrix each drug-indication pairs where was switched “unknown” one at time then recovered remaining pairs. Of probabilistically significant pairs, 70% successfully recovered. Next, applied whole narrow down candidates suggest alternative indications. able retrieve approved 6 drugs whose not listed SIDER. 908 present their indication information, suggested treatment options further investigations. Several uses can be supported from scientific literature. results demonstrated analyzed by generative model, discover associations between uses. In regard, serves as enrichment tool explore existing narrowing space.