Prediction of Protein Structural Class Based on Gapped-Dipeptides and a Recursive Feature Selection Approach

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Bioinformatics

Background:

• Understanding protein structural class is crucial for predicting protein function and tertiary structure.
• Accurate and efficient computational methods for protein structural class prediction remain a significant challenge in bioinformatics.
• Position-specific score matrix (PSSM) profiles offer valuable information for improving prediction accuracy but are underexplored.

Purpose of the Study:

• To develop an improved computational approach for predicting protein structural class.
• To leverage the information within PSSM profiles more effectively for enhanced prediction.
• To introduce a novel feature extraction and selection technique for this purpose.

Main Methods:

• A new feature extraction technique based on gapped-dipeptide composition computed directly from PSSM.
• Feature selection using Support Vector Machine-Recursive Feature Elimination (SVM-RFE) to identify optimal features.
• Construction of a final predictor using the selected optimal features.

Main Results:

• The proposed method successfully extracts informative features directly from PSSM.
• Jackknife tests on four datasets demonstrate satisfactory prediction accuracies.
• The feature extraction and selection strategy significantly improves protein structural class prediction.

Conclusions:

• The developed method provides a highly promising tool for predicting protein structural class.
• Feature extraction solely based on PSSM, combined with SVM-RFE, yields accurate predictions.
• This approach enhances the utility of PSSM profiles in bioinformatics predictions.