Pitfalls of Machine Learning in Membrane–permeability Prediction

Abstract

Regression–prediction models based on Fick’s law are popular approaches in prediction of membrane permeability rather than time–consuming molecular dynamics simulations of permeability process, but the prediction accuracies of these models remain insufficient for drug design, especially design of druggable macrocyclic molecules. In this review, we discuss the framework of mechanism–based regression model and modifications of the model based on the experiments and theoretical calculation.