Title:	AutoEncoder for SAXS reconstruction of Tr-WAXS data
DNr:	NAISS 2025/22-1083
Project Type:	NAISS Small Compute
Principal Investigator:	Leonardo Monrroy <leonardo.monrroy@kemi.uu.se>
Affiliation:	Uppsala universitet
Duration:	2025-08-15 – 2026-03-01
Classification:	10408
Keywords:

Abstract

We propose an AutoEncoder-based framework for reconstructing protein shapes from time-resolved SWAXS data. The model is trained on physically plausible protein deformations generated using AFsample2 and represented as 3D voxel models of proteins. By learning the deformation space, the trained network can be incorporated into a framework capable of generating candidate structures guided by fitting to time-resolved SWAXS profiles, enabling shape reconstruction directly from experimental measurements.