Angstrom AI

Our Science

Selected Publications

Computing hydration free energies of small molecules with first principles accuracy
JH Moore, DJ Cole, G Csanyi.
‍Preprint. 2024. Link
Expand
We introduce an efficient alchemical free energy method compatible with machine learned potentials , enabling, for the first time, calculations of biomolecular free energy with ab initio accuracy. Using a pretrained, transferrable, alchemically equipped MACE model, we demonstrate sub-chemical accuracy for the hydration free energies of organic molecules.
Flow Annealed Importance Sampling Bootstrap
LI Midgley, V Stimper, GNC Simm, B Schölkopf, JM Hernández-Lobato.
‍ICLR. 2023. Link
Expand
We propose FAB, a method for training generative models to sample from the Boltzmann distribution, using only the molecular energy function. We are the first to learn the Boltzmann distribution of the alanine dipeptide molecule using only the unnormalized target density, without access to samples generated via Molecular Dynamics (MD) simulations: FAB produces better results than training via maximum likelihood on MD samples while using 100 times fewer target evaluations. After reweighting the samples, we obtain unbiased histograms of dihedral angles that are almost identical to the ground truth.
SE(3) Equivariant Augmented Coupling Flows
LI Midgley, V Stimper, J Antorán, E Mathieu, B Schölkopf, JM Hernández-Lobato.
‍NeurIPS. 2023. Link
Expand
We propose an equivariant generative model architecture specialised to the task of estimating expectations with respect to molecular Boltzmann distributions. Our architecture is a discrete normalizing flow that is competitive with equivariant continuous flows and diffusion models, while allowing sampling more than an order of magnitude faster, and providing fast density evaluation.

Accelerating molecular simulation using generative AI

Our Science

Our Team

Interested in a demo?