DNABP: Identification of DNA-Binding Proteins Based on Feature Selection Using a Random Forest and Predicting Binding...

Area of Science:

• Molecular Biology
• Bioinformatics

Background:

• DNA-binding proteins are essential for numerous cellular functions.
• Predicting DNA-binding proteins from sequence data remains a challenge.
• Existing computational methods often require structural information, limiting their applicability.

Purpose of the Study:

• To develop a novel computational predictor, DNABP, for identifying DNA-binding proteins solely from their amino acid sequences.
• To introduce and evaluate a hybrid feature set incorporating physicochemical properties and residue binding propensities.
• To optimize the predictor through advanced feature selection techniques.

Main Methods:

• Utilized the random forest (RF) classifier.
• Developed a hybrid feature set combining novel sequence features related to physicochemical properties and binding propensities.
• Employed minimum redundancy maximum relevance (mRMR) and incremental feature selection (IFS) for feature optimization.

Main Results:

• The DNABP model achieved high prediction accuracy (86.90%), sensitivity (83.76%), and specificity (90.03%).
• The hybrid feature set significantly improved predictive performance compared to individual features.
• The Matthews correlation coefficient reached 0.727, indicating robust predictive power.

Conclusions:

• DNABP offers a reliable and accurate method for identifying DNA-binding proteins using only sequence information.
• The novel sequence features and optimized selection process contribute to the predictor's effectiveness.
• The freely available DNABP web server facilitates research in DNA-protein interactions.