Improving hybrid statistical and physical forcefields through local structure enumeration.

Area of Science:

• Computational biology
• Biophysics
• Structural bioinformatics

Background:

• Biomolecular simulations rely on force fields with physical and statistical energetic terms.
• Combining these terms presents challenges, particularly concerning the potential for double counting statistical and physical interactions.
• Existing force fields often exclude intra-residue physical terms to avoid overlap with statistical potentials.

Purpose of the Study:

• To develop a general scheme for correcting statistical potentials when used with physical terms in biomolecular force fields.
• To address the issue of double counting interactions in combined force field approaches.
• To enhance the accuracy and performance of protein structure prediction methods.

Main Methods:

• Developed a general scheme to correct statistical potentials in conjunction with physical terms.
• Applied these corrections to the sidechain torsional potential within the Rosetta all-atom force field.
• Identified and quantified double-counted electrostatic and steric interactions.

Main Results:

• The developed scheme successfully identified instances of double-counted interactions, such as sidechain-backbone electrostatics and Cβ-Cβ steric clashes.
• Enabled the re-inclusion of intra-residue physical terms that were previously excluded.
• Demonstrated improved performance in structure prediction tasks, including rotamer prediction and native structure discrimination.

Conclusions:

• The proposed correction scheme effectively resolves double counting issues when combining physical and statistical potentials.
• This approach leads to a more accurate and comprehensive biomolecular force field.
• The enhanced force field shows significant improvements in key protein structure prediction benchmarks.