BioCompLab focuses on the development of computational methods of machine learning, data mining and database techiniques to decipher and predict chemical/mixture bioactivity and toxicity. The developed methods can be utilized to accelerate the drug discovery and development, facilitate chemical/mixture risk assessment, reduce experimental animal uses and implement regulatory acceptable usage.
Chemical toxicity assessment is important for both drug development and risk assessment of environmental chemicals. We developed various approches using chemical/biological data to analyze and predict potential toxicity induced by chemicals/drugs/mixtures. Four major fields include 1) in silico toxicogenomics, 2) mixture effect, 3) virtual zebrafish, and 4) mechanism-based toxicity prediction. Online servers are available for the AI-based toxicity evaluation, e.g., SkinSensDB for AOP-based skin sensitization data , in silico toxicogenomics methods (ChemDIS) and machine learning models (SkinSensPred) to support the assessment of chemical/mixture toxicity.
Since the chemical bioactivity/toxicity data is scarce, limited applicability domain of models using conventional machine learning/deep learning algorithms is expected. We developed transfer learning/multitask learning approches to conquer the low data issues. Please refer to the published works of orphan GPCR drug discovery, skin sensitizers, pulmonary absoption for details.