神経機能を指標としたインビトロ毒性評価試験とAI

抄録

Micro-electrode array (MEA) assay using human iPSC-derived neurons are expected to one of in vitro assessment to predict the toxicity and predict the mechanism of action of chemical compounds. However, the analytical method that can predict the toxicity from MEA data are not established. In this study, we attempted to detect the risk ranking of pesticides from MEA data in cultured human iPSC-derived neurons. Human iPSC-derived neurons (Neucyte inc.) were cultured on Micro-electrode array (MEA) plate, and 15 pesticides were tested at 5 concentrations from 0.01 to 100μM. Using multivariate analysis of parameters for synchronized burst firings, we have succeeded in distinguishing the dose-dependent responses to pesticides into low, middle, and high risk. In addition, we found that deep learning using the divvied image data of raster plots can separate dose-dependent responses into low, middle, and high risk. Although there is a problem of in vitro to in vivo extrapolation, analytical methods using multivariate analysis and deep learning are useful for the detection of risk ranking of pesticides from MEA data in cultured hiPSC-derived neuronal networks.