Publication

Abstract : Protein secondary structure prediction is a critical task in bioinformatics, essential for understanding protein function and aiding drug discovery. Traditional methods have achieved significant milestones, but there remains a need for further accuracy and efficiency improvements. In this paper, we propose a novel approach that combines Bidirectional Long Short-Term Memory BiLSTM networks and Transformer models to enhance prediction accuracy. Our dual-model strategy leverages the sequential processing strengths of BiLSTM and the robust attention mechanisms of Transformers, enabling a more comprehensive analysis of protein sequences. Our extensive experiments demonstrate that this hybrid approach achieves superior performance, with Q3 accuracy reaching 92.03% and Q8 accuracy at 85.26%, significantly surpassing existing methods. This enhancement is ascribed to the capacity of BiLSTM to capture long-range dependencies and the Transformer's capability to focus on relevant parts of the sequence. The integration of these models provides a powerful tool for predicting protein secondary structures with high precision. The findings suggest that our approach not only advances the current state of PSSP but also offers valuable insights that can accelerate bioinformatics research and drug discovery processes. This work underscores the potential of combining deep learning models to tackle complex biological challenges more effectively