Enhanced Protein Secondary Structure Prediction with Bidirectional LSTM and Attention Mechanism

Abstract

Building an informational connection between the primary as well as tertiary structures of proteins depends on protein secondary structure. Protein Secondary Structure Prediction(PSSP) is a challenging task because coil states and protein properties do not clearly differ from one another. Furthermore, conducting manual trials on every single protein in the laboratory is costly and time-consuming. Computational protein structure prediction is a useful and promising field. An important application in the functional and structural research of proteins has been the precise prediction of the three state as well as eight state secondary structure of proteins. The findings of this study underscore the efficacy of employing bidirectional LSTM with attention mechanism on protein secondary structure datasets sourced from the RCSB Protein Data Bank. With accuracy rates of 91.11% and 85.07% for three state and eight state classifications respectively, the model outperforms both traditional bidirectional LSTM and bidirectional GRU architectures significantly. These results highlight the potential of leveraging attention mechanisms in conjunction with bidirectional LSTM networks for enhanced predictive performance in protein structure analysis. © 2024 IEEE.