A novel approach for protein secondary structure prediction using encoder-decoder with attention mechanism model

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Biology

Background:

• Protein secondary structure (PSS) prediction is a key challenge in computational biology.
• Accurate PSS prediction relies on understanding sequence-structure mapping and residue interactions.

Purpose of the Study:

• To develop an advanced computational model for protein secondary structure prediction.
• To leverage attention mechanisms for enhanced feature selection in sequence-structure mapping.

Main Methods:

• An encoder-decoder architecture incorporating an attention mechanism was proposed.
• The model was trained on the CB513 and CullPDB datasets.
• Performance was evaluated using Q3, Q8 accuracy, Segment of Overlap (SOV), and Mathew correlation coefficient.

Main Results:

• Achieved 70.63% Q3 and 78.93% Q8 accuracy on CullPDB.
• Attained 79.8% Q3 and 77.13% Q8 accuracy on CB513.
• Demonstrated SOV improvements up to 80.29% (CullPDB) and 91.3% (CB513).
• The model achieved high accuracy in very few training epochs, outperforming existing methods.

Conclusions:

• The proposed encoder-decoder model with attention effectively predicts protein secondary structures.
• The attention mechanism enhances the model's ability to capture critical residue interactions.
• This approach offers a computationally efficient and accurate solution for PSS prediction compared to state-of-the-art methods.