Improving protein secondary structure prediction based on short subsequences with local structure similarity

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Protein secondary structure (PSS) prediction is crucial for protein structural analysis and modeling.
• Current PSS prediction accuracy is limited, nearing an 80% plateau.
• Improving PSS prediction accuracy remains a significant challenge in bioinformatics.

Purpose of the Study:

• To develop an improved dictionary-based PSS prediction method, SymPred.
• To introduce a meta-predictor, SymPsiPred, for enhanced PSS prediction.
• To explore the application of natural language processing concepts for capturing sequence similarities.

Main Methods:

• Proposed the concept of "synonymous words" (n-gram patterns) to represent local sequence similarities and evolutionary variations.
• Developed a protein-dependent dictionary for PSS prediction using these synonymous words.
• Implemented SymPred and SymPsiPred, dictionary-based and meta-prediction tools, respectively.

Main Results:

• SymPred achieved an average Q3 accuracy of 81.0% on the DsspNr-25 dataset, outperforming existing methods.
• SymPsiPred reached an average Q3 accuracy of 83.9% on the DsspNr-25 dataset.
• SymPred demonstrated strong performance on independent test sets (EVA Set_1 and EVA_Set2) with Q3 accuracies of 78.8% and 79.2%.

Conclusions:

• Local sequence similarities in proteins correlate with conserved structures, enabling improved PSS prediction accuracy.
• Synonymous word analysis facilitates accurate alignment of structurally similar but evolutionarily distant protein sequences.
• The SymPred and SymPsiPred prediction servers are publicly available for research use.