EPDRNA

EPDRNA predicts DNA and RNA binding sites in disease-related proteins using an ensemble classifier to identify residue-level protein–nucleic acid interactions.

Key Features:

• Target scope: Predicts DNA and RNA binding residues in disease-related proteins at the sequence/residue level.
• Feature utilization: Uses Position-Specific Scoring Matrix (PSSM), physicochemical properties, and amino acid types as input features.
• Ensemble learning strategy: Integrates four machine learning algorithms into an ensemble with a soft voting mechanism.
• Training data sources: Model development and training utilize datasets sourced from UniProt and the Protein Data Bank (PDB).
• Evaluation protocol: Performance was assessed by 10-fold cross-validation on training datasets.
• Performance metrics: Achieved AUC values of 0.73 for DNA binding site prediction and 0.71 for RNA binding site prediction in 10-fold cross-validation.
• Independent test performance: On independent test datasets, reported recall and precision were 85% and 25% for protein–DNA interactions, and 82% and 27% for protein–RNA interactions.

Scientific Applications:

• Residue-level mapping: Identifying specific amino acid residues that interact with DNA or RNA in disease-related proteins.
• Molecular mechanism elucidation: Supporting studies of protein–nucleic acid interaction mechanisms relevant to human disease.
• Experimental prioritization: Prioritizing candidate binding residues for experimental validation and investigation of potential therapeutic targets.

Methodology:

Integrates four machine learning algorithms into an ensemble classifier with soft voting using PSSM, physicochemical properties, and amino acid types as input features, trained on UniProt and PDB datasets and evaluated by 10-fold cross-validation and independent test datasets.