Prediction of ADME properties, Part 1: Classification models to predict Caco-2 cell permeability using atom-based bilinear indices

Abstract

The prediction of the permeability through cultured Caco-2 cells (an often-used in vitro model for drug absorption) is carried out using theoretical models. Atom-based bilinear indices and linear discriminant analysis (LDA) are used to obtain quantitative models, which discriminate between higher absorption and moderate-poorer absorption compounds, form a database of measured PCaco-2 from a large data set with 157 structurally diverse compounds. We develop two LDA models with more than 90% of accuracy for training and test set; the best model presents accuracy of 91.79% and 91.30%, respectively. The results achieved in this work compare favourably with other approaches previously published in the technical literature. The percentage of good correlation was of 80%, in the virtual screening of 241 drugs with the reported values of the percentage of human intestinal absorption (HIA). Finally, we can say that, the present “in silico” method would be a valuable
tool in the drug discovery process in order to select the
molecules with the greatest chance before synthesis.