TargetDBP+: Enhancing the Performance of Identifying DNA-Binding Proteins via Weighted Convolutional Features

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Protein-DNA interactions are crucial for cellular processes.
• Accurate identification of DNA-binding proteins (DBPs) is essential for understanding these interactions.
• Current DBP identification methods require performance enhancement.

Purpose of the Study:

• To develop a novel and improved method for identifying DNA-binding proteins (DBPs).
• To enhance the accuracy and performance of DBP identification.
• To introduce a new benchmark dataset for evaluating DBP identification methods.

Main Methods:

• Extraction of five convolutional features: amino acid one-hot matrix (AAOHM), position-specific scoring matrix (PSSM), predicted secondary structure probability matrix (PSSPM), predicted solvent accessibility probability matrix (PSAPM), and predicted probabilities of DNA-binding sites (PPDBSs).
• Weighted serial combination of features using weights optimized by the differential evolution algorithm.
• Training the DBP identification model using the support vector machine (SVM) algorithm.
• Construction of a gold-standard benchmark dataset, UniSwiss, comprising 4881 DBPs and 4881 non-DBPs from UniprotKB/Swiss-Prot.

Main Results:

• TargetDBP+ achieved 85.83% accuracy and 88.45% precision on the UniSwiss validation subset.
• The method covered 82.41% of all DBP data.
• The Matthews correlation coefficient (MCC) value of 0.718 was significantly higher than those of other state-of-the-art methods.
• The UniSwiss dataset provides a robust benchmark for DBP identification.

Conclusions:

• TargetDBP+ represents a significant advancement in DBP identification.
• The developed method demonstrates superior performance compared to existing approaches.
• The UniSwiss dataset and TargetDBP+ web server/standalone program are valuable resources for the research community.