Machine learning approach in pharmacokinetics and toxicity prediction

Abstract

Quantitative-structure pharmacokinetic relationship (QSPkR) methods for predicting compounds of specific pharmacokinetic, or toxicological (ADMET) property are useful for facilitating drug discovery and drug safety evaluation. However QSPkR models were frequently built using datasets with a limited number of related compounds and by using linear statistical methods. Hence they may not be suitable for prediction of ADMET properties of diverse groups of compounds. Thus it is of interest to examine the potential of using a larger number and more diverse groups of compounds and non-linear machine learning methods, such as support vector machines and general regression neural network, in improving the quality of QSPkR models. The results show that QSPkR models developed in this work have higher prediction capabilities than those developed in earlier studies. A novel method for identification of relevant physicochemical and structural properties of a compound from non-linear QSPkR models, which are traditionally considered to be black boxes, is also introduced.