Applying Topic Modeling to Identify the Multifactorial Attributes of Drug-Induced Liver Injury

• Activity Number: 304 - Risk Applications for Disease, Toxicology, and Biomarker Modeling
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Section on Risk Analysis
• Author(s): Dale Bowman* and Ayako Suzuki and Jonathan Bona and Wen Zou and E. Olusegun George
• Companies: University of Memphis and Duke University School of Medicine and University of Arkansas for Medical Sciences and National Center for Toxicological Research and University of Memphis
• Keywords: drug-induced liver injury; Latent Dirichlet Application; pairwise mutual information

Abstract

Drug-induced liver injury (DILI) is one of the most common adverse reactions that results in early termination of drug development and withdrawal of drugs from the market. DILI is a complex disease with multiple contributing factors and interactions. Because of insufficient clinical knowledge, DILI is currently not predictable. Latent Dirichlet Allocation (LDA) has the potential to enhance biomedical investigation in multifactorial diseases such as DILI, where interacting factors may exert significant additive, synergetic, or antagonistic effects as a group although individual effects may be subtle. We used LDA as a Bayesian data-mining tool to investigate multifactorial determinants of DILI phenotypes, aiming to implement a new analytic approach for theory generation. More specifically, we sought to discover multifactorial determinants for chronic DILI using topic modeling and mutual information computed for 2, 3, 4, 5 and 6 factor combinations among identified topics and to generate theories for future investigation that can be validated in actual data sets.