Prediction of relative solvent accessibility by support vector regression and best-first method

Area of Science:

• Bioinformatics
• Computational Biology
• Protein Science

Background:

• Protein structure is intrinsically linked to its amino acid sequence.
• Mining large-scale protein sequence data presents challenges due to high dimensionality.
• Predicting protein properties like solvent accessibility is crucial for understanding structure-function relationships.

Purpose of the Study:

• To develop an efficient method for predicting relative solvent accessibility of proteins.
• To address the challenge of large datasets in bioinformatics by employing feature selection.
• To improve the accuracy and reduce computational time in protein property prediction.

Main Methods:

• A two-step procedure involving feature selection and support vector regression.
• Feature selection identifies a subset of evolutionary information based on physicochemical properties.
• Support vector regression utilizes selected features for predicting real-valued solvent accessibility.

Main Results:

• The proposed method demonstrates improved average prediction accuracy compared to existing approaches.
• Significant reduction in training time was observed.
• Effective identification of relevant evolutionary information for solvent accessibility prediction.

Conclusions:

• The presented two-step method offers a more accurate and computationally efficient approach for predicting protein solvent accessibility.
• This strategy effectively handles high-dimensional bioinformatics data.
• The findings contribute to advancing the analysis of protein sequences and structures.