Comparison of MLR, OPLS, and SVM as potent chemometric techniques used to estimate in vitro metabolic stability
Szymon Ulenberg, Mariusz Belka, Tomasz Bączek · Journal of Chemometrics · 2016
In the drug development pipeline, metabolic stability is not only one of the most limiting factors when it comes to drug candidate approval but is also one of the most resource‐consuming and time‐consuming ones regarding in vitro studies. With the popularization of the in silico approach in the drug discovery process and the application of many excellent chemometric techniques in predicting the properties of compounds, the authors focused on developing a model able to estimate the in vitro half‐life of arylpiperazine derivatives. The aim of the presented study was to develop a regression model and compare the performance of three potent methods (multiple linear regression (MLR), orthogonal partial least squares (OPLS), and support vector machine (SVM)) when predicting in vitro half‐life values. Of the three studied techniques, SVM proved to be the most accurate (RMSESVM = 0.054 for both sets, RMSEMLR = 0.072 for both sets, and RMSEOPLS = 0.073 for both sets) and provided the best predictability (Q2 = 0.8410; for MLR, Q2 = 0.7263; and for OPLS, Q2 = 0.7490). Hence, SVM regression provides valuable predictions and seems to be a promising technique not only for developing regression models but also for studying quantitative structure‐metabolic stability relationships. Copyright © 2016 John Wiley & Sons, Ltd.