Improved prediction of accessible surface area results in efficient energy function application

Area of Science:

• Bioinformatics and Computational Biology
• Structural Biology
• Protein Science

Background:

• Accessible Surface Area (ASA) is crucial for understanding protein structure and function.
• Accurate ASA prediction from protein sequence alone has broad applications in bioinformatics.
• Existing methods require improvement for enhanced utility in various predictive tasks.

Purpose of the Study:

• To develop a novel computational paradigm for accurate prediction of real-valued protein Accessible Surface Area (ASA) directly from amino acid sequence.
• To enhance the accuracy of ASA prediction for improved application in protein structure and function analysis.
• To integrate improved ASA predictions into energy functions for more effective protein modeling.

Main Methods:

• Developed REGAd(3)p, a predictor using Exact Regression with Regularization and a degree 3 polynomial kernel.
• Optimized the REGAd(3)p predictor using a Genetic Algorithm.
• Created and utilized a new benchmark dataset of 1001 protein chains for training and 298 for testing.
• Modeled prediction error as energy and integrated it with the 3DIGARS energy function to create 3DIGARS2.0.

Main Results:

• Achieved a maximum Pearson Correlation Coefficient (PCC) of 0.76 for ASA prediction.
• Demonstrated a 1.45% improvement in PCC over the existing top predictor, SPINE-X, on an independent test set.
• The resulting energy function, 3DIGARS2.0, significantly outperformed state-of-the-art energy functions on benchmark decoy sets.

Conclusions:

• The REGAd(3)p predictor offers a significant advancement in accurately estimating protein ASA from sequence.
• Improved ASA prediction directly benefits applications in protein structure prediction and function analysis.
• The novel 3DIGARS2.0 energy function, incorporating ASA prediction error, enhances protein structure modeling accuracy.