Modeling radical scavenging activity using molecular descriptors of unclassified compounds

Abstract

The overproduction of reactive oxygen species (ROS) leads to oxidative stress, which is associated with many human diseases. Antioxidants counteract the effects of ROS, but traditional assays are costly and time-consuming. Quantitative structure-activity relationship (QSAR) models offer a predictive alternative. We developed a QSAR model using data from 3133 unclassified antioxidant compounds using extreme gradient boosting (XGBoost), random forest (RF), and support vector machine (SVM) algorithms. Molecular descriptors were calculated using RDKit, and 82 were selected based on importance. The XGBoost model showed superior predictive performance, with good agreement with the experimental data (R-2 = 0.81). Descriptor analysis revealed a significant influence of phenolic groups on antioxidant activity. This research provides valuable insights for those wishing to predict antioxidant activity from unclassified compound structure data and has implications for industries such as drug discovery and efficacy evaluation. Through a large-scale analysis of 3133 unclassified antioxidant compounds, we present an advanced QSAR model covering a wide pIC50 range (-0.98-10.30). Unlike previous studies of restrictively classified compounds, we have achieved universality, which is expected to contribute to effective antioxidant activity prediction and candidate discovery.