Real value prediction of protein solvent accessibility using enhanced PSSM features.

Area of Science:

• Biochemistry
• Computational Biology
• Structural Biology

Background:

• Protein solvent accessibility (accessible surface area, ASA) prediction is crucial for tertiary structure determination.
• Traditional classification methods for ASA are limited by ill-defined states.
• Evolutionary information, like Position Specific Scoring Matrix (PSSM), is used for direct real-value ASA prediction.

Purpose of the Study:

• To enhance Position Specific Scoring Matrix (PSSM)-based features for accurate real-value accessible surface area (ASA) prediction.
• To develop a systematic method for identifying residue groups based on protein solvent accessibility.
• To improve the performance of ASA prediction by integrating physicochemical properties and amino acid solvent propensities.

Main Methods:

• A systematic method was developed to identify residue groups concerning protein solvent accessibility.
• Amino acid columns within the PSSM profile were merged to create novel features.
• Support Vector Regression (SVR) was employed to build a real-value ASA predictor.

Main Results:

• The proposed feature selection process generated informative features for ASA prediction.
• Experimental results demonstrated the effectiveness of the enhanced PSSM-based features.
• The developed predictor showed strong performance in predicting real-value ASA.

Conclusions:

• The proposed method achieved superior performance compared to existing ASA prediction packages on a benchmark dataset.
• The feature selection mechanism is applicable to other regression problems utilizing PSSM.
• The developed program and data are available upon request.