A novel feature selection method to predict protein structural class.

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Predicting protein structural class is crucial for understanding protein function.
• Integrating diverse protein properties enhances prediction accuracy but creates high-dimensional data challenges.
• Effective feature selection is essential for managing complex, integrated protein datasets.

Purpose of the Study:

• To propose a novel feature selection method, Partial-Maximum-Correlation-Information based Recursive Feature Elimination (PMCI-RFE).
• To enhance the prediction accuracy of protein structural class by selecting optimal feature subsets from high-dimensional integrated data.
• To enable the identification of informative features for analyzing biological relationships.

Main Methods:

• Developed the Partial-Maximum-Correlation-Information based Recursive Feature Elimination (PMCI-RFE) algorithm.
• Utilized correlation information between feature and class encoding spaces.
• Employed orthogonal component projection for feature space analysis.

Main Results:

• PMCI-RFE demonstrated fast and effective performance across six benchmark datasets.
• The method significantly improved protein structural class prediction accuracy compared to four state-of-the-art techniques.
• PMCI-RFE successfully leveraged information from various protein properties.

Conclusions:

• PMCI-RFE is a computationally efficient and effective feature selection tool.
• The method enhances the predictability of protein structural class by optimizing feature subsets.
• PMCI-RFE facilitates deeper analysis of biological relationships through informative feature identification.