AceFF: A State-of-the-Art Machine Learning Potential for Small Molecules
Stephen E. Farr, Stefan Doerr, Antonio Mirarchi, Francesc Sabanes Zariquiey, Gianni De Fabritiis
2026-01-07
Summary
This paper introduces AceFF, a new computer program designed to predict how molecules behave, specifically for finding new drugs.
What's the problem?
Currently, scientists use very accurate but slow computer simulations, like Density Functional Theory, to understand molecules. Faster methods exist using machine learning, but they often struggle to accurately predict behavior across a wide variety of different molecules you might find when searching for drugs. It's hard to create a 'one-size-fits-all' model that works well for everything.
What's the solution?
The researchers created AceFF, which is a machine learning model built using a specific architecture called TensorNet2. They trained it on a huge collection of data representing molecules that are likely to be useful as drugs. This training process allows AceFF to quickly and accurately predict how molecules will behave, balancing speed with the precision of those slower, more accurate methods. AceFF also handles molecules with electrical charges and includes all the common elements found in drugs like carbon, oxygen, and nitrogen.
Why it matters?
AceFF represents a significant improvement in the field because it's both fast and accurate. This means researchers can screen many more potential drug candidates much more quickly, potentially speeding up the drug discovery process and lowering costs. The program is also publicly available, allowing other scientists to use and build upon their work.
Abstract
We introduce AceFF, a pre-trained machine learning interatomic potential (MLIP) optimized for small molecule drug discovery. While MLIPs have emerged as efficient alternatives to Density Functional Theory (DFT), generalizability across diverse chemical spaces remains difficult. AceFF addresses this via a refined TensorNet2 architecture trained on a comprehensive dataset of drug-like compounds. This approach yields a force field that balances high-throughput inference speed with DFT-level accuracy. AceFF fully supports the essential medicinal chemistry elements (H, B, C, N, O, F, Si, P, S, Cl, Br, I) and is explicitly trained to handle charged states. Validation against rigorous benchmarks, including complex torsional energy scans, molecular dynamics trajectories, batched minimizations, and forces and anergy accuracy demonstrates that AceFF establishes a new state-of-the-art for organic molecules. The AceFF-2 model weights and inference code are available at https://huggingface.co/Acellera/AceFF-2.0.