Accurate Estimation of Solvent Accessible Surface Area for Coarse-Grained Biomolecular Structures with Deep Learning

Area of Science:

• Computational Biology
• Biophysics
• Machine Learning

Background:

• Coarse-grained (CG) models offer computational efficiency for biomolecular simulations.
• Accurate solvent accessible surface area (SASA) calculation is crucial for CG model applications like structure prediction and drug design.
• Existing SASA calculation methods struggle with CG biomolecular structures.

Purpose of the Study:

• To develop a deep learning-based method for accurate SASA estimation from CG biomolecular structures.
• To address the challenge of SASA calculation for CG protein and RNA models.

Main Methods:

• Developed DeepCGSA, a deep learning model for SASA estimation.
• Trained and validated DeepCGSA on various CG protein models (Cα-based, Cα-Cβ, Martini).
• Assessed performance on CG RNA and unfolded protein structures.

Main Results:

• DeepCGSA achieved near-perfect SASA estimation for CG protein models (correlation coefficients 0.95-0.99).
• Demonstrated significantly improved accuracy compared to existing methods.
• Showcased applicability and accuracy for CG RNA and unfolded protein structures.

Conclusions:

• DeepCGSA provides a highly accurate and efficient solution for SASA estimation in CG biomolecular models.
• The method is expected to significantly benefit protein/RNA structure prediction and drug design.
• DeepCGSA enhances the utility of CG models in computational biology and biophysics.