AMARO v2.0: one neural potential for proteins, ligands and complexes, 3.3x more pose recoveries
gdefabritiis · x · 2026-10-29
ACELLERA and Universitat Pompeu Fabra released a preprint for AMARO v2.0, a coarse-grained neural network potential covering proteins, drug-like ligands and their complexes, including ternary protein–protein–ligand systems.
Design
- Retains every heavy atom; hydrogens encoded into 26 bead types spanning 18 elements, conditioned on total system charge, implicit solvent
- Training goes from quantum-mechanical small-molecule forces to protein conformations to protein–ligand complexes
Validation
- Two 500 ns ubiquitin simulations preserve the fold; captures α3D and villin conformational landscapes, improving α3D over v1.0
- Adapting to complexes reduces isolated-molecule force accuracy (an acknowledged trade-off)
Binding highlight: in trypsin–benzamidine tests from unbound starts, AMARO recovered the crystal pose in 30/100 trajectories (4 GPU-hours each) vs 9/100 for all-atom molecular mechanics — 3.3x more recoveries despite lower simulated-time throughput.
More from Research
- Preprint: Human peer reviewers caught only 27% of major errors, all 15 AI models beat them (53-96%) — anshulkundaje · 2026-10-29
- New paper rules out Hamiltonian Phase States as a route to quantum-only Microcrypt — StefanoGogioso · 2026-10-29
- MSR AI for Science team hiring postdoc to generate molecular data for Skala — FrankNoeBerlin · 2026-10-29
- ML-enhanced resonant soft X-ray scattering enables sub-monolayer tracking of buried-interface molecular states — carbon-lab · 2026-10-29
- Any3R: one feedforward model handles 13 camera types for 3D reconstruction, beating OmniVGGT and PanoVGGT — zhenjun_zhao · 2026-10-29
- GeoGSplat: geometry priors top DTU and NRGBD with just 65K Gaussians — zhenjun_zhao · 2026-10-29