Valorizing Food Ingredients Using Machine Learning Models

The classification of peptides, particularly antibacterial peptides, is a critical task in bioinformatics with applications in drug discovery and therapeutic development. This study explores the use of Graph Neural Networks (GNNs), specifically Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs), to classify peptides based on their functional properties. Unlike traditional sequence-based approaches, we incorporate structural information by constructing peptide graphs from 3D structures generated using AlphaFold. We evaluate multiple feature representations, including one-hot encoding and physicochemical properties, to assess their impact on model performance. Experimental results indicate that GCN and GAT models can effectively classify peptides, with GCN achieving higher accuracy but exhibiting overfitting tendencies. The integration of additional features contributed to model stability but did not lead to substantial performance gains. Our findings suggest that combining structural insights with deep learning can enhance peptide classification, but limitations such as the availability of experimentally validated 3D structures and sequence redundancy remain challenges. Future work should focus on refining graph representations, incorporating advanced transformer-based models, and improving computational 3D structure predictions. Enhancing dataset diversity and leveraging hybrid architectures may further improve classification accuracy and generalization.