Protein side chain modeling with orientation-dependent atomic force fields derived by series expansions.

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• Accurate protein side chain modeling is crucial for understanding protein function and dynamics.
• Existing force fields often struggle to precisely predict side chain conformations, limiting their application in protein structure refinement.

Purpose of the Study:

• To develop novel, high-accuracy force fields for protein side chain modeling.
• To enhance the prediction of native side chain conformations using optimized energy functions.

Main Methods:

• Developed distance-dependent (OSCAR-d) and orientation-dependent (OSCAR-o) energy functions.
• Optimized 802 parameters for OSCAR-d and 1087 parameters for OSCAR-o using extensive training datasets.
• Validated OSCAR-o on 218 protein structures, comparing performance against established methods.

Main Results:

• OSCAR-o achieved 88.8% accuracy for chi(1) and 79.7% for chi(1+2) dihedral angles.
• Demonstrated superior performance with an overall RMSD of 1.24 Å, compared to 1.40 Å for the next-best method.
• Achieved high accuracy for core residues (0.62 Å RMSD), indicating robust modeling of protein interiors.

Conclusions:

• The developed OSCAR force fields provide accurate and efficient protein side chain modeling.
• Continuous energy functions are suitable for gradient-based optimization in protein structure refinement.
• A publicly available program for protein side chain prediction using OSCAR is provided.