hybridDBRpred

hybridDBRpred predicts DNA-binding residues in protein sequences by integrating predictions from structure-trained and disorder-trained models to improve accuracy across both structured protein–DNA complexes and intrinsically disordered proteins.

Key Features:

• Integration of Diverse Predictors: Leverages the strengths of three top-performing DBR prediction tools and incorporates predictions from ten existing DBR predictors identified by empirical analysis.
• Deep Transformer Meta-model: Employs a deep transformer network to integrate and synthesize predictions from constituent predictors.
• Cross-annotation Performance: Achieves high accuracy for DNA-binding residue prediction across structured protein–DNA complexes and intrinsically disordered proteins.
• Reduction of Cross-predictions: Reduces erroneous cross-predictions of non-DNA ligand-interacting residues that are common in disorder-trained predictors.
• Statistical Superiority: Empirical analyses show it outperforms each of the ten individual tools and baseline meta-predictors such as averaging and logistic regression.

Scientific Applications:

• Gene regulation studies: Maps DNA-binding residues relevant to gene regulatory mechanisms.
• Transcription factor binding analysis: Identifies potential transcription factor DNA-binding residues and binding sites.
• Mechanistic studies of protein–DNA interactions: Supports investigation of molecular mechanisms underlying protein–DNA interactions.
• Functional annotation of novel proteins: Identifies potential DNA-binding regions in novel or poorly characterized proteins.

Methodology:

Performed empirical analysis of ten existing DBR predictors to identify top performers for structured and disordered annotations, and integrated those predictions using a deep transformer network as a meta-model.