PiRaNhA

PiRaNhA predicts RNA-binding residues in protein sequences using support vector machines to support functional annotation of RNA-binding proteins.

Key Features:

• Support Vector Machine: An SVM model predicts RNA-binding residues using sequence-derived features.
• Input features: The SVM is trained on position-specific scoring matrices (PSSMs), residue interface propensity, predicted residue accessibility, and residue hydrophobicity.
• Residue-level performance: On a non-redundant set of 42 protein sequences not used in training it achieved 85% accuracy, 90% specificity, and a Matthews correlation coefficient (MCC) of 0.41, and on 81 RNA-binding proteins using 5-fold cross-validation it attained 87.2% accuracy and MCC 0.50.
• Protein-level prediction: Decision values from the residue-level SVM are used as input to a secondary SVM to predict RNA-binding function at the protein level, achieving MCC 0.53 and accuracy 76.1%.

Scientific Applications:

• Functional annotation of proteins: Predicts RNA-binding residues to assist annotation of RNA-binding function in proteins.
• Experimental design: Provides residue-level predictions to guide targeted experimental verification of RNA–protein interactions.
• Proteome-wide annotation: Enables proteome-scale prediction of RNA-binding proteins via protein-level SVM classification.

Methodology:

An SVM is trained on PSSMs, residue interface propensity, predicted residue accessibility, and residue hydrophobicity; decision values feed a secondary SVM for protein-level classification, and performance was evaluated on a non-redundant set of 42 sequences and on 81 RNA-binding proteins using 5-fold cross-validation.