Prediction of novel and selective TNF-alpha converting enzyme (TACE) inhibitors and characterization of correlative molecular descriptors by machine learning approaches

• Published in: Journal of molecular graphics & modelling Vol. 28; no. 3; pp. 236 - 244
• Main Authors: Cong, Yong, Yang, Xue-gang, Lv, Wei, Xue, Ying
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.10.2009
• Subjects: ADAM Proteins - antagonists & inhibitors; ADAM17 Protein; Algorithms; Artificial Intelligence; Enzyme Inhibitors - analysis; Humans; Inhibitor; Machine learning (ML); Non-inhibitor; The support vector machine (SVM); TNF-α converting enzyme (TACE); Machine learning (ML); Inhibitor; TNF-α converting enzyme (TACE); Non-inhibitor; The support vector machine (SVM)
• ISSN: 1093-3263; 1873-4243
• DOI: 10.1016/j.jmgm.2009.08.001

The inhibition of TNF-α converting enzyme (TACE) has been explored as a feasible therapy for the treatment of rheumatoid arthritis (RA) and Crohn's disease (CD). Recently, large numbers of novel and selective TACE inhibitors have been reported. It is desirable to develop machine learning (ML) models for identifying the inhibitors of TACE in the early drug design phase and test the prediction capabilities of these ML models. This work evaluated four ML methods, support vector machine (SVM), k-nearest neighbor ( k-NN), back-propagation neural network (BPNN) and C4.5 decision tree (C4.5 DT), which were trained and tested by using a diverse set of 443 TACE inhibitors and 759 non-inhibitors. A well-established feature selection method, the recursive feature elimination (RFE) method, was used to select the most appropriate descriptors for classification from a large pool of descriptors, and two evaluation methods, 5-fold cross-validation and independent evaluation, were used to assess the performances of these developed models. In this study, all these ML models have already achieved promising prediction accuracies. By using the RFE method, the prediction accuracies are further improved. In k-NN, the model gives the best prediction for TACE inhibitors (98.32%), and the SVM bears the best prediction for non-inhibitors (99.51%). Both the k-NN and SVM model give the best overall prediction accuracy (98.45%). To the best of our knowledge, the SVM model developed in this work is the first one for the classification prediction of TACE inhibitors with a broad applicability domain. Our study suggests that ML methods, particularly SVM, are potentially useful for facilitating the discovery of TACE inhibitors and for exhibiting the molecular descriptors associated with TACE inhibitors.