SVM-Cabins: prediction of solvent accessibility using accumulation cutoff set and support vector machine

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Protein Chemistry

Background:

• Predicting solvent accessibility (ASA) is crucial for understanding protein structure and function.
• Existing methods often predict discrete states or real values of ASA, with limitations in converting between them.
• Integrating discrete and real-valued ASA prediction approaches remains a challenge.

Purpose of the Study:

• To develop a novel method for predicting numerical real values of solvent accessibility.
• To integrate discrete state prediction with real-valued ASA prediction for improved accuracy.
• To establish a robust method for estimating accessible surface area (ASA) in amino acid residues.

Main Methods:

• Utilized a support vector machine (SVM) combined with an accumulation cutoff set (SVM-Cabins).
• Predicted discrete states of ASA from amino acid residue evolutionary profiles.
• Mapped predicted discrete ASA states onto a real-valued linear space using algebraic methods.

Main Results:

• Achieved a mean absolute error of 15.1% for real-valued ASA prediction.
• Demonstrated a coefficient of correlation equal to 0.66 on a dataset of 502 proteins.
• Showcased performance comparable to the best existing ASA prediction methods.

Conclusions:

• The SVM-Cabins method effectively predicts real-valued solvent accessibility by first predicting discrete ASA states.
• This approach optimizes prediction performance for both binary and real-valued states simultaneously.
• The method offers a rigorous and accurate way to estimate accessible surface area (ASA).