Jun 2026· Nature Communications· Vol 17· 0 citations· 95 references
Medicine
TL;DR
In tests covering drug-like molecules, molecular dimers, torsional energy profiles, energy-minimum structures, and biomolecular simulations, Residual Learning Force Field gives accurate and stable predictions across diverse systems.
Abstract
Accurate descriptions of interactions between atoms are essential for molecular simulations used to study biology and support drug discovery. Existing force fields often face a trade-off between physical reliability, computational efficiency, and accuracy across unfamiliar molecules. Here we show that Residual Learning Force Field, a hybrid machine learning force field, can reduce this trade-off by combining simple physics-based descriptions of bonded interactions with learned corrections for remaining energetic effects. The two components are trained together through a three-step strategy so that each contributes complementary information. In tests covering drug-like molecules, molecular dimers, torsional energy profiles, energy-minimum structures, and biomolecular simulations, Residual Learning Force Field gives accurate and stable predictions across diverse systems. These results suggest that combining physical constraints with data-driven corrections can provide a practical route toward more reliable and efficient molecular simulation for biological research and drug discovery. Molecular force fields underpin biomolecular simulation but often trade accuracy for generality. This study presents ResFF, a residual-learning model that combines physical rules with neural corrections to improve accuracy while retaining generality.
GeoNet is a physicochemical-principle-guided framework for modeling dual-range atomic interactions that achieves the smallest model size and the shortest training time, demonstrating both superior predictive performance and computational efficiency.
Machine-learned force fields (MLFFs) contain many learnable parameters and therefore require large training datasets. This poses a challenge for developing highly accurate, general-purpose MLFFs because generating high-quality ab initio reference data is computationally expensive. Classical empirical potentials offer a potentially inexpensive source of synthetic training data, but existing models often lack the accuracy needed to provide useful reference energies. Here, we introduce the density-based intermolecular potential (DensIP), a physics-based model of intermolecular interactions that uses machine-learned electron densities and only four universal parameters. We train and test DensIP on CCSD(T)/CBS interaction energies from DES15K, a dataset of dimers of small organic molecules. DensIP achieves sub-kcal/mol errors for dimers containing molecules absent from the training set, including molecules in non-equilibrium conformations, demonstrating strong transferability. We further show that DensIP can be applied to molecules as large as drug ligands. Notably, DensIP outperforms state-of-the-art general-purpose MLFFs for long-range interactions, making it a promising approach for generating accurate synthetic training data at scale.
Dahvyd Wing, Mihail Bogojeski, Szabolcs Góger et al.· 0 citations
The quantitative prediction of biomolecular recognition is crucial to molecular science. The challenge is not merely structural determination but the prediction of (thermo)dynamic and kinetic observables arising from high-dimensional molecular ensembles, such as free energies, conformational distributions, and rate processes across different conditions. As the field shifts from structure-centric to ensemble-based descriptions, two complementary modeling strategies have matured: explicit energy-based approaches grounded in statistical mechanics and data-driven models that learn statistical representations of molecular configurations from large data sets. Physics-based methods, including molecular dynamics and free energy perturbation, estimate observables by sampling (Boltzmann-distributed) configurations under approximate molecular Hamiltonians, thereby providing mechanistic interpretability and thermodynamic consistency, albeit at non-negligible computational cost and with inherent force field limitations. In contrast, modern machine learning approaches rapidly generate structures and propose conformational ensembles without explicit thermodynamic weighting, by learning statistical patterns in structural and bioactivity data. While these methods often achieve high predictive performance, they do not inherently enforce thermodynamic consistency due to the lack of an explicit connection to a partition function and thus may produce configurations that are not physically realizable. We argue that, since physics-based simulations and machine learning provide complementary approximations to the underlying probability distribution associated with biomolecular recognition events, and they excel respectively in consistency with free-energy landscapes and state populations and in predictive accuracy, the central challenge for the coming decade will be integrating them into hybrid frameworks that are scalable and transferable.
R. Khalil, Elena Frasnetti, Han Kurt et al.· Journal of Physical Chemistr...· 0 citations
A machine learning approach is presented that accelerates DFTB simulations by predicting optimal initial atomic charges and demonstrates that ML-predicted initial charges consistently and significantly improve SCC convergence across diverse chemical systems including organic molecules, biomolecules, water clusters, transition metal oxides and solid electrolytes.
Maximilian L. Ach, Karsten Reuter, C. Panosetti· 0 citations
Molecular dynamics (MD) is essential for investigating atomic‐scale processes in materials and molecular systems, but the cost of high‐accuracy machine learning force field simulations still limits accessible system sizes and timescales. Here, we propose a practical model‐switching strategy for Deep Potential (DP)‐based MD simulations that alternates between independently trained DP models with different cutoff radii: a standard 6 Å model for higher accuracy and a reduced‐cutoff 4 Å model for faster inference. The method was implemented in LAMMPS/DeePMD and evaluated using solid‐phase anatase TiO
2
and liquid‐phase polyethylene glycol (PEG). For anatase TiO
2
, the 1:3 4–6 Å switching scheme preserved radial distribution function (RDF) correlations of 0.996 or higher relative to the 6 Å baseline while achieving a 1.24‐fold speedup. For PEG, the switching scheme maintained RDF correlations of 0.996 or higher with a 1.18‐fold speedup. Additional optimization using network‐size reduction and mixed‐precision inference achieved a 2.53‐fold speedup with RDF correlations of 0.975–0.988. Constant particle‐number, pressure, and temperature (NPT) simulations remained stable, whereas constant particle‐number, volume, and energy (NVE) simulations revealed system‐dependent energy‐drift behavior, particularly for aggressively optimized models. These results demonstrate that DP model switching provides a simple and practical route for accelerating structural MD simulations while highlighting the need for validation when strict energy conservation is required.
Ryuya Kanda, Megumu Yamazaki, Yuta Yoshimoto et al.· Advanced Intelligent Discove...· 0 citations
We introduce implicit machine learning force fields (I-MLFFs), which replace explicit stacks of neural network layers with self-consistent fixed-point equations. In molecular simulations, this formulation enables intermediate representations to be reused across successive timesteps, thereby warm-starting force evaluation. The resulting models effectively combine the computational footprint of a shallow, single-layer MLFF with the representational capacity and accuracy of a deep neural network. Our approach unlocks architecture-agnostic efficiency gains that are inaccessible when force prediction and trajectory integration are considered separately. We demonstrate this across three major classes of graph neural networks: invariant, equivariant Cartesian tensor, and SO(3)-equivariant spherical-tensor architectures. Each yields a two- to five-fold reduction in compute and memory footprint. Crucially, these gains are achieved while retaining full atomistic resolution and the original integration timestep, avoiding spatial or temporal coarse graining. Our contribution therefore advances the scaling frontier of quantum-mechanically faithful molecular simulation, enabling longer trajectories and larger atomistic systems within fixed GPU memory and compute budgets, and thereby opening access to new insights across biomolecular and material systems.
J. Maess, Leon Werner, J. Frank et al.· 0 citations