Bioinformatics-based identification of assays that inform on disease hazard

Scott Auerbach, Raymond Tice and Kristine Thayer, National Toxicology Program at NIEHS, USA

Animal-based hazard characterization, although notably comprehensive in its coverage of disease space, is incapable of detecting disease hazards where animal models are underdeveloped (e.g., autism). In order to identify disease-centric molecular targets that can be developed into high throughput screens for the detection of disease-specific hazard, we have performed a bioinformatic analysis that employs multi-database/gene-centric literature mining, gene class information, and expression. First disease-gene associations were mined from a number of resources including the Comparative Toxicogenomics Database, Ingenuity Pathways Knowledgebase, GeneGo Knowledgebase, OMIM, Entrez Gene, CoPub, GeneCards, and Phenopedia. Associated genes were then filtered based on whether they encode a protein with a ‘druggable domain’ meaning there is a high probability that they will interact with small molecules. Using this approach, we have identified screening targets for autism (GLO1, OXTR, GABRA5, SLC6A4, CHRNA4), Type 1 diabetes (CTLA4, PTPN22, IL2RA, ITPR3), Type 2 diabetes (PPARG, KCNJ11, HNF4A, ABCC8) and obesity (PPARG, PPARA, SERPINE1, PPARD). Gene-linked assay data from phase 1 of the EPA’s ToxCast program is used in combination with our bioinformatic-based analysis of disease to identify hypothetical relationships between chemical exposure and disease hazard.