Protein function annotation from sequence: prediction of residues interacting with RNA

Area of Science:

• Bioinformatics
• Computational Biology
• Protein Function Annotation

Background:

• Eukaryotic proteomes contain many proteins with unknown functions.
• Computational methods are crucial for initial protein function annotation.
• Predicting RNA-binding residues (RBRs) is key to understanding protein function.

Purpose of the Study:

• To develop and evaluate PiRaNhA, a method for predicting RBRs from protein sequence.
• To assess PiRaNhA's utility for predicting RNA-binding function at the protein level.
• To facilitate targeted experimental annotation of proteins.

Main Methods:

• Trained a support vector machine (SVM) using sequence properties: position-specific scoring matrices, interface propensities, predicted accessibility, and hydrophobicity.
• Evaluated PiRaNhA using 5-fold cross-validation on known RNA-binding proteins.
• Utilized PiRaNhA decision values with a second SVM for protein-level RNA-binding function prediction.

Main Results:

• PiRaNhA achieved a Matthews Correlation Coefficient (MCC) of 0.50 and 87.2% accuracy in cross-validation.
• On unseen proteins, PiRaNhA yielded an MCC of 0.41 and 84.5% accuracy.
• Protein-level RNA-binding function prediction achieved an MCC of 0.53 and 76.1% accuracy.

Conclusions:

• PiRaNhA effectively predicts RNA-binding residues and protein-level RNA-binding function.
• The method aids in targeted experimental design for protein function annotation.
• PiRaNhA shows promise for large-scale, proteome-wide functional annotations.