TargetDBP: Accurate DNA-Binding Protein Prediction Via Sequence-Based Multi-View Feature Learning.

Area of Science:

• Computational Biology
• Bioinformatics
• Protein Science

Background:

• Identifying DNA-binding proteins (DBPs) is crucial for protein function annotation.
• Accurate DBP identification from sequence data remains a significant computational challenge.

Purpose of the Study:

• To develop a novel computational method, TargetDBP, for accurate DBP identification from protein sequences.
• To improve the accuracy of predicting DNA-binding proteins.

Main Methods:

• Feature extraction using Amino Acid Composition (AAC), Pseudo Position-Specific Scoring Matrix (PsePSSM), Pseudo Predicted Relative Solvent Accessibility (PsePRSA), and Pseudo Predicted Probabilities of DNA-Binding Sites (PsePPDBS).
• Weight optimization using a differential evolution algorithm.
• Feature selection via Support Vector Machine Recursive Feature Elimination with Correlation Bias Reduction (SVM-REF+CBR).
• Model training using Support Vector Machine (SVM).

Main Results:

• TargetDBP achieved higher performance compared to existing state-of-the-art predictors on a newly constructed benchmark dataset.
• The method effectively integrates multiple sequence-derived features for DBP prediction.
• A gold-standard, non-redundant benchmark dataset was created from the PDB database.

Conclusions:

• TargetDBP presents a robust and accurate computational approach for identifying DNA-binding proteins.
• The developed method enhances protein function annotation capabilities.
• The TargetDBP web server and datasets are publicly available for academic research.