Deep Learning for Protein Structure Prediction: Advancements in Structural Bioinformatics

Abstract Motivation Accurate prediction of protein structures is crucial for understanding protein function, stability, and interactions, with far-reaching implications in drug discovery and protein engineering. As the fields of structural bioinformatics and artificial intelligence continue to converge, a standardized model for protein structure prediction is still yet to be seen as even large models like AlphaFold continue to change architectures. To this end, we provide a comprehensive literature review highlighting the latest advancements and challenges in deep learning-based structure prediction, as well as a benchmark system for structure prediction and visualization of amino acid protein sequences. Results We present ProteiNN, a Transformer-based model for end-to-end single-sequence protein structure prediction, motivated by the need for accurate and efficient methods to decipher protein structures and their roles in biological processes and a system to perform prediction on user-input protein sequences. The model leverages the transformer architecture’s powerful representation learning capabilities to predict protein secondary and tertiary structures directly from integer-encoded amino acid sequences. Our results demonstrate that ProteiNN is effective in predicting secondary structures, though further improvements are necessary to enhance the model’s performance in predicting higher-level structures. This work thus showcases the potential of transformer-based architectures in structure prediction and lays the foundation for future research in structural bioinformatics and related fields.