Advancing in Protein Function Prediction: Leveraging Small Data Techniques in Convolutional and Graph Convolutional Networks

Protein function prediction is a crucial task in computational biology and bioinformatics, aiding in understanding cellular processes and potential therapeutic targets. In this study, we present a comprehensive comparative analysis of two advanced deep learning approaches for protein function prediction: Convolutional Neural Networks (CNNs) and Graph Convolutional Networks (GCNs). Specifically, we leverage Deep-FRI, a state-of-the-art structure-based function prediction model employing Graph Convolutional Networks, and Lite-SeqCNN+, a lightweight deep CNN architecture tailored for protein function prediction. Our investigation involves a detailed examination of the strengths and weaknesses of both approaches, considering factors such as predictive accuracy, computational efficiency, and interpretability. We evaluate the models on benchmark datasets like CAFA and explore their performance across diverse protein families. The comparative analysis sheds light on the nuances of CNNs and GCNs in the context of protein function prediction, providing valuable insights for researchers and practitioners in the field of computational biology.