Host: The Japanese Society of Toxicology
Drugs are small compounds that target specific proteins and induce a perturbation of the entire intracellular biomolecular pathway. While drug administration has the expected effect, it can also cause toxicity and side effects. The molecular mechanism of toxicity and side effects remains largely unidentified. In recent years, a large amount of various information on drugs and compounds has been accumulated in publicly available databases. Such information includes various data such as chemical structure, target protein, drug efficacy, toxicity, side effects, etc., and it is expected that the relationship between them will be clarified by a computer approach. We propose a computational method for associating information on different scales, that is, information on the molecular level of compound-target protein interaction and information on the phenotypic level of compound toxicity / side effect information. We analyzed the correlation between target protein interaction data of compounds and toxicity / side effect data of compounds using machine learning models such as sparse canonical correlation analysis and sparse logistic regression. As a result of protein enrichment analysis of correlation sets using molecular pathway information, proteins working in similar molecular pathways were clustered in the same correlation set even if their molecular functions were different. That is, it can be interpreted that the functional control of proteins clustered in a certain correlation set by a compound may cause similar side effects through the activation and inactivation of the same molecular pathway. The proposed method is expected to be useful for predicting the potential toxicity / side effects of a compound from the target protein profile of the compound, as well as enabling estimation of the molecular mechanism of action of toxicity / side effects.
