TNF i pred: a classification model to predict TNF-α inhibitors

Rheumatoid arthritis (RA), characterized by severe inflammation in the joint lining, is a progressive, chronic, autoimmune disorder with high morbidity and mortality rates. There are several mechanisms responsible for joint damage, but overproduction of TNF-α is a significant mechanism that results in excess swelling and pain. Drugs acting on TNF-α are known to significantly reduce the disease progression and improve the quality of life in many RA patients. Hence, inhibiting TNF-α is considered one of the most effective treatments for RA. Currently, there are only a few FDA-approved TNF-α inhibitors, which are mainly monoclonal antibodies, fusion proteins, or biosimilars with disadvantages such as poor stability, difficulty in route of administration (often given as injection or infusion), cost-prohibitive large-scale production, and increased side effects. There are just a handful of small compounds known to have TNF- inhibitory capabilities. Thus, there is a dire need for new drugs, especially small molecules in the market, such as TNF-α inhibitors. The conventional method of identifying TNF-α inhibitors is expensive, labor, and time intensive. Machine learning (ML) can be used to solve existing drug discovery and development problems. In this study, four classification algorithms—naïve Bayes (NB), random forest (RF), k-nearest neighbor (kNN), and support vector machine (SVM)—were used to train ML models for classifying TNF-α inhibitors based on three sets of features. The performance of the RF model was found to be best when using 1D, 2D, and fingerprints as features, with an accuracy of 87.96 and a sensitivity of 86.17. To our knowledge, this is the first ML model for TNF-α inhibitor prediction. The model is available at http://14.139.57.41/tnfipred/