Abstract:Machine learning potentials have emerged as a means to enhance the accuracy of biomolecular simulations. However, their application is constrained by the significant computational cost arising from the vast number of parameters compared to traditional molecular mechanics. To tackle this issue, we introduce an optimized implementation of the hybrid method (NNP/MM), which combines neural network potentials (NNP) and molecular mechanics (MM). This approach models a portion of the system, such as a small molecule, using NNP while employing MM for the remaining system to boost efficiency. By conducting molecular dynamics (MD) simulations on various protein-ligand complexes and metadynamics (MTD) simulations on a ligand, we showcase the capabilities of our implementation of NNP/MM. It has enabled us to increase the simulation speed by 5 times and achieve a combined sampling of one microsecond for each complex, marking the longest simulations ever reported for this class of simulation.

What problem does this paper attempt to address?

The paper mainly addresses the following issues: 1. **Enhancing the speed of molecular dynamics simulations**: By combining neural network potentials (NNP) with molecular mechanics (MM), an optimized hybrid method (NNP/MM) was developed to improve the speed of molecular dynamics simulations for small molecule systems, including drug-like molecules. 2. **Improving simulation accuracy**: Utilizing the advantages of neural network potentials in enhancing the accuracy of biomolecular simulations, especially when dealing with novel drug-like molecules that are difficult to parameterize accurately in traditional force fields. 3. **Accelerating computational efficiency**: To address the low computational efficiency of the NNP/MM method, researchers proposed a series of optimization strategies, including migrating all computational tasks to the GPU, implementing a customized CUDA kernel for the ANI-2x feature extractor, and parallelizing the neural network computation process, thereby significantly improving the computational speed. 4. **Validating the method's effectiveness**: Through a series of molecular dynamics (MD) and metadynamics (MTD) simulation experiments, the effectiveness and practicality of the optimized NNP/MM method were validated. These experiments covered the simulation of drug fragments and four protein-ligand complexes, achieving longer simulation time scales than previously reported. 5. **Performance improvement**: The optimized NNP/MM method increased the simulation speed by approximately 5 times. Although it is still slower compared to traditional MM methods, this progress significantly narrows the performance gap between the two and leaves room for further optimization. In summary, this study aims to overcome the computational bottlenecks of neural network potentials in large-scale molecular dynamics simulation applications by optimizing the NNP/MM method, while maintaining or even improving simulation accuracy, providing a powerful tool for fields such as drug discovery.

NNP/MM: Accelerating molecular dynamics simulations with machine learning potentials and molecular mechanic

NNP/MM: Accelerating Molecular Dynamics Simulations with Machine Learning Potentials and Molecular Mechanics

Machine Learning for Parameter Auto-tuning in Molecular Dynamics Simulations: Efficient Dynamics of Ions near Polarizable Nanoparticles

OpenMM 8: Molecular Dynamics Simulation with Machine Learning Potentials

Molecular Dynamics with Neural-Network Potentials

Accurate and efficient molecular dynamics based on machine learning and non von Neumann architecture

Ab-Initio Molecular Dynamics Acceleration Scheme with an Adaptive Machine Learning Framework

Simulate Time-integrated Coarse-grained Molecular Dynamics with Geometric Machine Learning

Accelerated Quantum Mechanics/Molecular Mechanics Simulations Via Neural Networks Incorporated with Mechanical Embedding Scheme.

Machine Learning for Performance Enhancement of Molecular Dynamics Simulations

Pushing the limit of molecular dynamics with ab initio accuracy to 100 million atoms with machine learning

Machine Learning in QM/MM Molecular Dynamics Simulations of Condensed-Phase Systems

Accurate Free Energy Calculation via Multiscale Simulations Driven by Hybrid Machine Learning and Molecular Mechanics Potentials

Accelerating ab initio QM/MM Molecular Dynamics Simulations with Multiple Time Step Integration and a Recalibrated Semi-empirical QM/MM Hamiltonian

Scaling Molecular Dynamics with ab initio Accuracy to 149 Nanoseconds per Day

Training One DeePMD Model in Minutes: a Step Towards Online Learning.

Accelerating Metadynamics-Based Free-Energy Calculations with Adaptive Machine Learning Potentials.

Accurate interatomic force field for molecular dynamics simulation by hybridizing classical and machine learning potentials

A machine learning-oriented pseudo-field approach to accelerate runtime of molecular dynamics simulation of liquids

Graph Neural Networks Accelerated Molecular Dynamics