Machine Learning-Based Prediction of Biological Activity in Natural Product Phytochemicals

Abstract

Natural product phytochemicals hold significant promise for drug discovery due to their diverse biological activities and potential to offer novel therapeutic solutions. These compounds are derived from plants, fungi, and other natural sources, providing a rich source of chemical diversity that is often not found in synthetic libraries. Their complex structures and unique mechanisms of action can lead to the development of drugs with new and innovative therapeutic properties. However, discovering and developing new drugs from these compounds presents challenges due to the reliance on knowledge-based methods, which can be time-consuming and inefficient. To address this issue, we propose a novel bioactivity classification network designed to predict the biological activities of phytochemical samples. This study utilized a dataset categorized into four groups: Antioxidant, Toxicity, Anti-inflammatory and Immune, and Lipid Metabolism. The performance of the network was assessed using key metrics including Accuracy, Recall, Precision, and F1-Score. The results revealed that a model with a depth of 4 layers achieved the highest performance. The proposed network achieved an accuracy of 0.7926 and an F1-Score of 0.7248 on the public dataset.