From Structure to Function: Biological Activity Prediction of Phytochemicals Using Molecular Fingerprints with Convolutional Neural Networks

Phytochemicals, naturally occurring compounds in plants, offer significant potential for drug development due to their diverse structures and biological activities. They exhibit antioxidant, anti-inflammatory, antimicrobial, anticancer, cardiovascular, and neuroprotective properties, making them beneficial for treating various health conditions. Advantages of phytochemicals include their natural origin, better safety profiles, multi-targeted actions, synergistic effects with other compounds, and sustainability. However, the traditional knowledge-based approach to analizing phytochemicals is limited by its reliance on well-known, locally available plants, resulting in a narrow scope and frequent redundancy. This approach often depends on anecdotal and subjective evidence, faces the risk of knowledge loss, and encounters ethical and legal issues. To address this issue, I propose a convolutional neural network-based systematic approach to predict potential biological activities from molecular structure inputs. The proposed system converts molecular structures into one-hot vector representations using SMILES notation and molecular fingerprint algorithms. These vectors are then fed into a biological activity prediction network to estimate possible biological activities. Through comprehensive experiments, I have demonstrated that applying a convolutional neural network-based machine learning approach yields promising results by achieving an accuracy of 87.8%.