InFlamPred: a machine learning framework for anti-inflammatory small molecule prediction

Abstract

Several anti-inflammatory small molecules have been discovered and utilized in the treatment of various inflammatory and autoimmune diseases. Despite experimental identification of numerous anti-inflammatory peptides, the development of peptide-based drugs remains expensive, time-consuming, and labor-intensive. Small molecules offer higher stability compared to peptides, owing to their chemical synthesis or natural sources. Consequently, there is an urgent need to develop advanced machine learning (ML) methods utilizing a large-scale dataset consisting of experimentally acquired small molecule to enhance precision and efficiency. This study introduces a predictive ML-method, named InFlamPred (Anti-inflammatory Small Molecule Predictor), tailored for anti-inflammatory small molecules. The proposed ML classifier facilitates compound screening in inflammatory diseases. We trained five different ML classifiers—RF, KNN, LGBM, DT, and XGB achieving an overall accuracy ranging from 61 to 75%. Notably, the LGBM, RF, and XGB models demonstrated strong performance on the training dataset, with XGB maintaining high accuracy and robustness on the test dataset, achieving an accuracy of 75% and an AUC of 81%. When evaluated on an external validation dataset consisting of 80 compounds (equally distributed between active and inactive classes), the XGB model effectively identified all active and inactive molecules. These results highlight the reliability and generalizability of the proposed ML approach in accurately predicting small molecules with anti-inflammatory potential.