XSSDense: Time-resolved X-ray Solution Scattering Density Reconstruction Using a Variational Autoencoder
TL;DR
XSSDense is proposed, a framework that couples a variational autoencoder trained on electron densities from predicted or simulated protein ensembles with a genetic algorithm to refine densities against scattering data and enables structurally grounded electron density reconstructions that intrinsically capture conformational heterogeneity.
Abstract
Solution X-ray scattering provides unique structural information on biomolecules under biological conditions, resolving conformational heterogeneity and time-resolved structural changes. The scattering profiles contain limited information, and interpretation largely relies on fitting candidate structures guided by priors. Direct reconstruction of electron density maps is desirable, but so far has been prevented by the difficulty of incorporating such prior knowledge. Here we propose XSSDense, a framework that couples a variational autoencoder trained on electron densities from predicted or simulated protein ensembles with a genetic algorithm to refine densities against scattering data. We validate XSSDense on synthetic data for crambin, recover the conformational heterogeneity of the unfolded state of Avena sativa light-oxygen-voltage sensing domain 2, resolve a de-novo density for the pre-unfolding state of the same protein, and provide a new structural description of the signalling-state ensemble of photoactive yellow protein. XSSDense enables structurally grounded electron density reconstructions that intrinsically capture conformational heterogeneity.