Active Learning for Machine Learning Driven Molecular Dynamics

2025-09-21

AI Summary

Machine learning–driven coarse-grained molecular dynamics (CG-MD) suffers from potential energy degradation and poor generalization in unexplored conformational regions. Method: This paper proposes an active learning–based online optimization framework that integrates an RMSD-driven frame selection strategy with the CGSchNet neural network potential model. It dynamically identifies gaps in conformational space coverage and queries a full-atom simulation “oracle” to generate high-fidelity training data on-the-fly. Contribution/Results: Compared to static training paradigms, the framework significantly enhances model self-correction capability and computational efficiency. Experiments on the Chignolin protein show a 33.05% reduction in Wasserstein-1 distance within tICA-embedded space relative to baseline methods, demonstrating improved exploration and modeling accuracy for unseen conformations. This work establishes a scalable, adaptive paradigm for CG-MD simulation.

Technology Category

• Machine Learning: Active Learning
• Search and Optimization: Sampling/Simulation-based Search
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

Abstract

Machine learned coarse grained (CG) potentials are fast, but degrade over time when simulations reach undersampled biomolecular conformations, and generating widespread all atom (AA) data to combat this is computationally infeasible. We propose a novel active learning framework for CG neural network potentials in molecular dynamics (MD). Building on the CGSchNet model, our method employs root mean squared deviation (RMSD) based frame selection from MD simulations in order to generate data on the fly by querying an oracle during the training of a neural network potential. This framework preserves CG level efficiency while correcting the model at precise, RMSD identified coverage gaps. By training CGSchNet, a coarse grained neural network potential, we empirically show that our framework explores previously unseen configurations and trains the model on unexplored regions of conformational space. Our active learning framework enables a CGSchNet model trained on the Chignolin protein to achieve a 33.05% improvement in the Wasserstein 1 (W1) metric in Time lagged Independent Component Analysis (TICA) space on an in house benchmark suite.

Problem

• Addressing degradation of machine learned coarse-grained potentials in undersampled biomolecular conformations
• Developing active learning framework to generate training data efficiently during molecular dynamics simulations
• Improving neural network potential accuracy by targeting unexplored regions of conformational space

Innovation

• Active learning framework for CG neural network potentials
• RMSD-based frame selection from MD simulations
• On-the-fly data generation by querying oracle during training