A feature-based approach to predict hot spots in protein-DNA binding interfaces

Area of Science:

• Computational biology
• Bioinformatics
• Structural biology

Background:

• Protein-DNA binding hot spots are key to binding energy.
• Experimental hot spot identification is costly and slow.
• Computational prediction methods are needed for large-scale analysis.

Purpose of the Study:

• To develop and evaluate computational methods for predicting protein-DNA binding hot spots.
• To identify optimal features for hot spot prediction.
• To create a user-friendly web server for hot spot prediction.

Main Methods:

• Systematic assessment of 114 features from sequence, structure, network, and solvent accessibility.
• Training and comparison of machine learning models: SVM, random forest, Naïve Bayes, k-NN.
• Development of the PrPDH method using SVM and 10 selected optimal features.

Main Results:

• Solvent accessible surface area features significantly impact prediction.
• Combining complementary features enhances prediction performance.
• Support Vector Machine (SVM) demonstrated superior performance over other models.
• The PrPDH predictor achieved better performance than state-of-the-art methods.

Conclusions:

• PrPDH provides an effective computational approach for predicting protein-DNA binding hot spots.
• The developed method offers improved accuracy compared to existing predictors.
• A freely available web server facilitates the application of PrPDH in research.