From Sequence to Structure: Transformer based Secondary Structure and GNN for Tertiary Structure Prediction

In computational biology, it is still quite difficult to predict the tertiary structure of proteins based on their amino acid sequence. In order to accurately predict the tertiary level, this paper proposes a new method that takes advantage of transformer models and graph neural networks (GNN). The directed approach extracts features from the protein sequence and captures long-range dependencies using a transformer-based model. After that, these features are fed into a GNN, which simulates the local interactions among amino acid residues by exploiting the natural graph structure of proteins. Detailed understanding of the structural and sequence information contained in protein sequences is possible with this combined approach. The CASP11 dataset, a standard for evaluating protein structure prediction techniques, is used to evaluate the correctness of the method. The proposed method’s results predicts tertiary protein structures with encouraging accuracy, which could make it a useful tool to improve protein structure prediction and deepen our knowledge of protein function.