MolecularDiffusion.modules.models.goflow¶
GoFlow: transition-state geometry from a reaction’s condensed graph, via conditional flow matching and an E(3)-equivariant transformer.
Galustian, Mark, Karwounopoulos, Kovar & Heid, GoFlow: Efficient
Transition State Geometry Prediction with Flow Matching and E(3)-Equivariant
Neural Networks, ChemRxiv (2025), doi:10.26434/chemrxiv-2025-bk2rh. Ported
from the repo checked out at others/nice/goflow (commit
3ec00a09d9b283e3258ae01fe5d3e35bb3812bff).
The package holds the network (gotennet, ops,
outputs, cgr_graph_utils) and the flow-matching algorithm
(flow) only. Its data layer lives in
data/component/goflow_data.py (beside the shared reaction_data.py
container) and its task in modules/tasks/diffusion_goflow.py.
See docs/model_integrations/goflow/INTEGRATION_PLAN.md.