Enhancing Biomolecular Simulations with Hybrid Potentials Incorporating NMR Data.

Area of Science:

• Biomolecular Simulation
• Computational Chemistry
• Structural Biology

Background:

• Hybrid restraint potentials combine experimental data with molecular mechanics force fields.
• These potentials enhance biomolecular simulation methods like molecular dynamics and structure prediction.
• Accurate force fields are crucial for reliable computational predictions.

Purpose of the Study:

• To develop a novel hybrid restraint potential using the NapShift neural network.
• To leverage NapShift's differentiable architecture for protein structural refinement.
• To assess the performance of the NapShift hybrid potential in global optimization and molecular dynamics.

Main Methods:

• Developed a hybrid restraint potential integrating NapShift (a neural network for protein NMR chemical shift prediction) with molecular mechanics force fields.
• Utilized NapShift's differentiability to compute energy penalties and forces based on predicted vs. experimental chemical shifts.
• Benchmarked the hybrid potential using basin-hopping global optimization and molecular dynamics simulations.

Main Results:

• The NapShift hybrid potential significantly improved the accuracy of biomolecular simulations.
• Enhanced structure prediction accuracy was observed in basin-hopping simulations.
• Increased local stability was achieved in molecular dynamics simulations using the NapShift hybrid potential.

Conclusions:

• Neural network-based hybrid potentials using NMR observables can enhance various molecular simulation techniques.
• The accuracy of these potentials is expected to increase with more experimental data.
• This approach offers a promising avenue for improving protein structure prediction and dynamics.