Prediction of Protein Structural Classes for Low-Similarity Sequences Based on Consensus Sequence and Segmented PSSM

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Bioinformatics

Background:

• Predicting protein structural classes is crucial for understanding protein functions, regulation, and interactions, especially for low-similarity sequences.
• Feature extraction from protein sequences, including primary sequence, predicted secondary structure, and Position-Specific Scoring Matrix (PSSM), is key to accurate prediction.
• PSSM-based prediction methods have significantly improved accuracy in recent years.

Purpose of the Study:

• To propose a novel computational method, CSP-SegPseP-SegACP, for predicting protein structural classes.
• To enhance prediction accuracy for low-similarity protein sequences by integrating multiple feature types.
• To evaluate the proposed method's performance against existing PSSM-based approaches.

Main Methods:

• Developed CSP-SegPseP-SegACP by fusing consensus sequence (CS), segmented PsePSSM, and segmented autocovariance transformation (ACT) features derived from PSSM.
• Constructed a 700-dimensional feature vector, reduced to 224 dimensions using Principal Component Analysis (PCA).
• Validated the method using three standard low-similarity datasets (1189, 25PDB, 640) and jackknife cross-validation.

Main Results:

• The CSP-SegPseP-SegACP method demonstrated favorable and competitive performance on the tested datasets.
• Rigorous jackknife cross-validation confirmed the method's effectiveness.
• The proposed approach showed significant improvements compared to existing PSSM-based prediction methods.

Conclusions:

• CSP-SegPseP-SegACP offers a powerful new tool for predicting protein structural classes, particularly for challenging low-similarity sequences.
• The fusion of CS, segmented PsePSSM, and ACT features provides a robust feature representation.
• This method serves as a valuable complement to existing PSSM-based techniques in structural bioinformatics.