Performance of universal machine-learned potentials with explicit long-range interactions in biomolecular simulations

2025-08-14

AI Summary

This study addresses three key challenges hindering the application of general-purpose machine learning interatomic potentials (MLIPs) in biomolecular simulation: inadequate modeling of long-range interactions, imbalanced training data distributions, and insufficient evaluation frameworks. We systematically assess the suitability of equivariant message-passing neural networks (EGNNs) for protein modeling. Using the SPICE-v2 dataset, we construct a multi-scale EGNN model that explicitly incorporates dispersion and electrostatic long-range contributions, and design targeted conformational sampling strategies alongside error-decomposition-based evaluation metrics. Our results demonstrate that: (1) explicit electrostatic terms significantly enhance conformational diversity in Trp-cage simulations; (2) predictive performance depends critically on training data composition—not merely model size; and (3) current benchmark protocols inadequately capture biologically relevant properties, including conformational distributions and thermodynamic stability. This work provides both methodological advances—such as physically informed long-range coupling and refined evaluation—and empirical evidence to support reliable MLIP deployment in protein simulations.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Multiagent Systems: Multiagent Learning

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

Abstract

Universal machine-learned potentials promise transferable accuracy across compositional and vibrational degrees of freedom, yet their application to biomolecular simulations remains underexplored. This work systematically evaluates equivariant message-passing architectures trained on the SPICE-v2 dataset with and without explicit long-range dispersion and electrostatics. We assess the impact of model size, training data composition, and electrostatic treatment across in- and out-of-distribution benchmark datasets, as well as molecular simulations of bulk liquid water, aqueous NaCl solutions, and biomolecules, including alanine tripeptide, the mini-protein Trp-cage, and Crambin. While larger models improve accuracy on benchmark datasets, this trend does not consistently extend to properties obtained from simulations. Predicted properties also depend on the composition of the training dataset. Long-range electrostatics show no systematic impact across systems. However, for Trp-cage, their inclusion yields increased conformational variability. Our results suggest that imbalanced datasets and immature evaluation practices currently challenge the applicability of universal machine-learned potentials to biomolecular simulations.

Problem

• Evaluating universal machine-learned potentials in biomolecular simulations
• Assessing impact of model size and training data composition
• Challenges in applying universal potentials to biomolecular systems

Innovation

• Equivariant message-passing architectures for biomolecular simulations
• Explicit long-range dispersion and electrostatics integration
• Training on SPICE-v2 dataset with varied compositions