Classification of Cytochrome P450 Inhibitors and Noninhibitors Using Combined Classifiers

Abstract

Adverse side effects of drug-drug interactions induced by human cytochrome P450 (CYP) inhibition is an important consideration, especially, during the research phase of drug discovery. It is highly desirable to develop computational models that can predict the inhibitive effect of a compound against a specific CYP isoform. In this study, inhibitor predicting models were developed for five major CYP isoforms, namely 1A2, 2C9, 2C19, 2D6, and 3A4, using a combined classifier algorithm on a large data set containing more than 24,700 unique compounds, extracted from PubChem. The combined classifiers algorithm is an ensemble of different independent machine learning classifiers including support vector machine, C4.5 decision tree, k-nearest neighbor, and naïve Bayes, fused by a back-propagation artificial neural network (BP-ANN). All developed models were validated by 5-fold cross-validation and a diverse validation set composed of about 9000 diverse unique compounds. The range of the area under the receiver operating characteristic curve (AUC) for the validation sets was 0.764 to 0.815 for CYP1A2, 0.837 to 0.861 for CYP2C9, 0.793 to 0.842 for CYP2C19, 0.839 to 0.886 for CYP2D6, and 0.754 to 0.790 for CYP3A4, respectively, using the new developed combined classifiers. The overall performance of the combined classifiers fused by BP-ANN was superior to that of three classic fusion techniques (Mean, Maximum, and Multiply). The chemical spaces of data sets were explored by multidimensional scaling plots, and the use of applicability domain improved the prediction accuracies of models. In addition, some representative substructure fragments differentiating CYP inhibitors and noninhibitors were characterized by the substructure fragment analysis. These classification models are applicable for virtual screening of the five major CYP isoforms inhibitors or can be used as simple filters of potential chemicals in drug discovery.

Activity-Limiting Role of Molecular Size: Size-Dependency of Maximum Activity for P450 Inhibition as Revealed by qHTS Data

“…P450s were already the subject of many QSAR studies (Ekins et al, 1999;Lewis et al, 2001;Riley et al, 2001;Wanchana et al, 2003;Hansch et al, 2004;Mao et al, 2006;Gleeson et al, 2007;Yamashita et al, 2008Yamashita et al, , 2011 as well as inhibitor/noninhibitor classification studies (Jensen et al, 2007;Choi et al, 2009;Cheng et al, 2011). For large ADME/Tox characterization panels, it has been repeatedly found that molecular size plays a major, defining role.…”

Recent progresses in the exploration of machine learning methods as in-silico ADME prediction tools

“…predicting ADME regulatory properties ML classification methods have also been extensively used for predicting regulators of drug ADME properties, particularly the inhibitors of drug efflux and influx transporters for regulating multi-drug resistance ( [13,15,16], improvement of predictive performance by such strategies as the use of expanded training datasets [8,[64][65][66][67], and both objectives [13,16]. The SEs, SPs, and ACs in predicting P-glycoprotein inhibitors are in ranges of 58%-99%, 47%-91%, and 62%-94% respectively.…”

“…The combined classifier approach, illustrated in Figure 9, explores the collective predictive power of multiple ML classifiers, with the probability output of each independent ML model processed by a decision network [13]. This decision network consists of two layers of units, an input layer and an output layer.…”

“…These probability outputs are used as new descriptors to develop a neural network model, such as a Back-Propagation ANN model [13], that generates the final combination decision probability (P C +1 and P C -1…”

“…In particular, the combined ML models can be developed such that multiple ML models are fused by a neural network platform, which have been shown to generate highly robust predictive models with less sensitivity on the choices of the ML models and parameters [13].…”