Sovereign 8s.is mirror of standardgalactic/molexa
  • Python 74.7%
  • Jupyter Notebook 25.3%
Find a file
Repository files (latest commit first)
Filename Latest commit message Latest commit date
2025-12-22 14:46:21 -08:00
mnodes add mnodes train 2025-01-12 01:26:23 -08:00
model selectively freeze deoniser and evaluator 2024-09-26 13:35:04 -07:00
notebooks update example notebook 2025-02-02 23:57:52 -08:00
train add mnodes train 2025-01-12 01:26:23 -08:00
utils update example notebook 2025-02-02 23:57:52 -08:00
environment.yml Update environment.yml 2025-12-22 14:46:21 -08:00
LICENSE Create LICENSE 2025-07-29 12:02:21 -07:00
README.md Update README.md 2025-12-22 14:45:02 -08:00

MOLEXA

Fig_1_annotated

MOLEXA is a deep generative neural network capable of reconstructing molecular geometries from ion momentum measurements in X-ray-based Coulomb explosion imaging (CEI) experiments. It is built on the Transformer architecture and the diffusion generative modeling framework. It allows for inversion of momentum-space datasets to position space, thus providing the structure of a molecule right before its interaction with an X-ray pulse. In addition, it can provide an uncertainty estimate for its reconstructed molecular geometries. By employing time-resolved CEI datasets, MOLEXA is able to provide ”snapshots” of a molecule at different instants during a chemical reaction. This enables the use of the CEI technique for direct reconstruction of molecular dynamics as they unfold on femtosecond or longer time scales.



Main architecture

Dynamics Extraction Module
image


Structure Denoising Module
image




Exemplary reconstructions

The predicted and ground-truth structures are shown in opaque and semi-transparent colors, respectively.

image



Acknowledgements

The training framework is based on Fastai's miniai library.