Quantum machine learning corrects classical forcefields: Stretching DNA base pairs in explicit solvent.

Area of Science:

• Computational chemistry
• Molecular dynamics
• Biophysics

Background:

• Classical forcefields in molecular dynamics (MD) simulations often lack accuracy for complex systems like DNA.
• Discrepancies exist between experimental DNA stretching behavior and classical simulation predictions.

Purpose of the Study:

• To improve the accuracy of MD simulations by incorporating quantum mechanical data.
• To investigate the mechanical properties of DNA under tension and resolve simulation-model discrepancies.

Main Methods:

• Developed a kernel-based machine learning method trained on quantum-mechanical fragment energies.
• Applied the method to generate a potential-energy surface for a DNA duplex, including solvation and electron exchange-correlation effects.
• Introduced Kernel Modified Molecular Dynamics (KMMD) as a generalizable approach for biomolecular simulations.

Main Results:

• Classical DNA models exhibit excessive stiffness in stretching compared to experimental data.
• The quantum correction qualitatively aligns simulation results with experimental thermodynamics for larger DNA helices.
• Identified a potential explanation for the long-standing discrepancy between DNA stretching experiments and classical simulations.

Conclusions:

• KMMD offers a promising approach to enhance the accuracy of molecular simulations.
• The developed quantum correction method addresses a key limitation in current DNA modeling.
• KMMD is now available within the AMBER22 simulation software, facilitating broader application in biomolecular research.