SREPRHot

SREPRHot predicts binding hotspots in protein-RNA interactions by identifying residues whose alanine mutation yields a change in binding free energy (ΔΔG) ≥ 2.0 kcal/mol to support analysis of molecular recognition and therapeutic targeting.

Key Features:

• Definition of Binding Hotspots: Binding hotspots are defined as residues whose mutation to alanine results in a binding free energy change (ΔΔG) ≥ 2.0 kcal/mol.
• Data Balancing with SMOTE: The Synthetic Minority Over-sampling Technique (SMOTE) is used to generate synthetic minority-class samples to address dataset imbalance.
• Residue Interface Propensity: A novel feature that captures the propensity of residues to participate in protein-RNA interfaces.
• Topological Features: Features are derived from node-weighted networks to provide structural and functional insights into residues within protein-RNA complexes.
• Random Grouping Feature Selection Strategy: A random grouping strategy combined with a two-step method is applied to determine an optimal feature subset.
• Ensemble Modeling Approach: A stacking ensemble classifier integrates multiple model predictions to improve accuracy and robustness.
• Performance on Independent Test: Reported performance on an independent testing dataset is Sensitivity (SEN) 0.900, Matthews Correlation Coefficient (MCC) 0.557, and Area Under the Curve (AUC) 0.829.

Scientific Applications:

• Structural biology: Identifying binding hotspots to inform interpretation of protein-RNA complex structures.
• Molecular biology: Highlighting functionally critical residues for studies of molecular recognition and mutational effects.
• Drug discovery: Prioritizing residue targets for therapeutic design that modulates protein-RNA interactions.
• Computational biology: Enabling large-scale in silico analysis of protein-RNA interaction hotspots using machine-learning features and ensemble models.

Methodology:

Datasets are balanced using SMOTE; conventional and novel features including residue interface propensity and topological characteristics from node-weighted networks are extracted; a random grouping feature selection strategy with a two-step method refines the feature set; prediction is performed using a stacking ensemble classifier.