分子軌道法とニューラルネットワークを基盤とした薬物吸収予測システムの開発

点击翻译按钮获取中文摘要

Drug discovery and development of drugs has been time- and labor-intensitve. A key issue to solve this problem is to develop knowledge bases such as design of drug molecules and formulations. Permeability of biological membranes are important factors determining bioavailability and effectiveness of drugs. In this context, we developed mathematical models for predicting them from molecular structures of compounds. Structural descriptors of drugs were obtained by molecular orbital calculations, and the relationship between the descriptors and permeability coefficients through human skin was analyzed by multiple linear regression or artificial neural network. We revealed that dipole moment, polarizability, and atomic charges of drug molecules are important descriptors to predict skin permeability for drugs, and that a neural network model is much superior to a linear regression model in describing the descriptor/permeability relationship. This approach was applicable to prediction of permeability of Caco-2 cell monolayers, that have been used as a model of intestinal epithelium. Thus, the combination of molecular orbital calculation and neural network might be useful to predict pharmacokinetic properties of drugs. In addition, we analyzed the effect of vehicles on skin permeability of drugs based on a two-layer skin diffusion model, in which the skin is composed of stratum corneum and the lower layer, and found a close relationship between skin permeability and dielectric constant of vehicles. In conclusion, this study might be able to accelerate effectiveness of drug discovery and development, especially for topical formulations.