Host: The Japanese Society of Toxicology
Name : The 50th Annual Meeting of the Japanese Society of Toxicology
Date : June 19, 2023 - June 21, 2023
When we humans think of the properties of a chemical, for example, a chemical has functional groups A and B, has toxicity T, and has a molecular weight of M, we recognize it as such. Although we can handle only a few dimensions in this way, we can manage higher dimensional information with a computer. Thus, for expansion of our recognition for a target with a computer, it is an indispensable operation to quantify the target as high-dimensional information comprehensively and without arbitrariness. Modeling is also indispensable for outputting the information of high-dimensionality thus expanded to a dimension that we can recognize again. In other words, numericalization and modeling enable us to handle information beyond our perception and to discover new aspects of the object of analysis.
Our group is developing research on the understanding and utilization of chemicals based on the above-mentioned principle of "Expansion with Computer and Extraction with Modeling". For instance, we have taken the approach of describing the effects of chemicals by capturing the biological responses of cultured cells treated with chemicals using transcriptome data, and of adapting a latent variable model to them for finding the common mechanisms behind the effects of the chemicals. End-to-end analysis, in which general deep learning models excel, is a sequential operation consisting of feature extraction and modeling, and is located in the same framework.
In this presentation, we focus on how to represent chemicals for toxicological studies and safety assessment. The presentation will focus on the recent progress in decomposition of chemical effects, numerization of chemical structures, and numerization of chemical effects in vivo.
