Prediction of catalytic residues using Support Vector Machine with selected protein sequence and structural...

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• The rapid increase in protein sequences from genome projects outpaces functional characterization.
• Predicting protein function is crucial, with catalytic sites offering key insights.
• Existing computational methods for catalytic site prediction suffer from low accuracy and high false positive rates.

Purpose of the Study:

• To develop a novel, accurate computational method for predicting catalytic sites in proteins.
• To identify optimal machine learning algorithms and discriminative features for catalytic residue prediction.

Main Methods:

• Compared 26 machine learning classifiers using a dataset of 79 enzymes and 254 catalytic residues.
• Employed Sequential Minimal Optimization (SMO), a Support Vector Machine (SVM), as the best-performing algorithm.
• Utilized Wrapper Subset Selection to identify an optimal set of seven predictive residue properties.

Main Results:

• The SMO algorithm achieved over 86% predictive accuracy, correctly identifying 228 out of 254 catalytic residues.
• The selected optimal subset included sequence conservation, amino acid catalytic propensities, and surface position.
• The method demonstrated a low false negative rate, missing only 10.2% of catalytic residues.

Conclusions:

• The developed method accurately predicts catalytic residues, offering a valuable tool for functional annotation.
• This computational approach can serve as a "catalytic residue filter" to guide experimental validation.
• The findings facilitate the functional characterization of proteins with known structures but unknown functions.