MLIPAudit: A benchmarking tool for Machine Learned Interatomic Potentials

TL;DR

MLIPAudit provides a unified, open benchmarking framework for machine-learned interatomic potentials, spanning general MD stability, small-molecule energetics, molecular liquids, and biomolecular sampling. It introduces a modular benchmark suite with pre-computed baselines and a transparent, composite scoring system, plus an online leaderboard to compare models on downstream tasks. The work demonstrates cross-domain transferability of several state-of-the-art MLIPs while highlighting domain-specific gaps and the need for broader data coverage and longer-timescale evaluations. By enabling easy running, comparison, and contribution, MLIPAudit aims to accelerate robust, reproducible development of MLIPs for complex molecular systems.

Abstract

Machine-learned interatomic potentials (MLIPs) promise to significantly advance atomistic simulations by delivering quantum-level accuracy for large molecular systems at a fraction of the computational cost of traditional electronic structure methods. While model hubs and categorisation efforts have emerged in recent years, it remains difficult to consistently discover, compare, and apply these models across diverse scenarios. The field still lacks a standardised and comprehensive framework for evaluating MLIP performance. We introduce MLIPAudit, an open, curated and modular benchmarking suite designed to assess the accuracy of MLIP models across a variety of application tasks. MLIPAudit offers a diverse collection of benchmark systems, including small organic compounds, molecular liquids, proteins and flexible peptides, along with pre-computed results for a range of pre-trained and published models. MLIPAudit also provides tools for users to evaluate their models using the same standardised pipeline. A continuously updated leaderboard tracks performance across benchmarks, enabling direct comparison on downstream tasks. By providing a unified, transparent reference framework for model validation and comparison, MLIPAudit aims to foster reproducibility, transparency, and community-driven progress in the development of MLIPs for complex molecular systems. In order to illustrate the use of the library, we present some benchmarks run on a series of internal models, along with publicly available ones (UMA-Small, MACE-OFF, MACE-MP). The library is available on GitHub at https://github.com/instadeepai/mlipaudit, on PyPI at https://pypi.org/project/mlipaudit/ under the Apache License 2.0, and the leaderboard can be accessed on HuggingFace at https://huggingface.co/spaces/InstaDeepAI/mlipaudit-leaderboard.

Figures (6)

• Figure 1: Representative molecular systems spanning increasing levels of structural and environmental complexity, from isolated dimers and drug-like molecules, to condensed-phase molecular liquids and folded biomolecules.
• Figure 2: Reactivity benchmark performance. (a–b) Reaction energy profiles for two Grambow reactions (IDs 008805 and 000433) grambow MLIP predictions to DFT references. (c) MAEs for activation energies (EA) and reaction enthalpies across the benchmark.
• Figure 3: Water radial distribution function for the example models, compared with the experimental observable and two water classical forcefields TIP3P and TIP4P tip3p
• Figure 4: Dihedral scan benchmark. (a) Dihedral energy profiles for fragment 015 compared to DFT reference values. (b) MAE and RMSE for each model. DFT-level error threshold (red dashed line).
• Figure 5: Conformer selection benchmark across three pharmaceutically relevant molecules: adenosine (ADO), benzylpenicillin (BPN), and efavirenz (EFA). MAE is computed with respect to DFT reference conformer energies. DFT threshold (red dashed line at 0.5 kcal/mol). Insets depict representative 3D conformers for each molecule.