Automated matching of two-time X-ray photon correlation maps from protein dynamics with Cahn-Hilliard type simulations using autoencoder networks

TL;DR

This paper presents an automated machine learning approach using autoencoders to classify X-ray photon correlation maps from protein solutions, integrating simulations and experimental data for efficient analysis.

Contribution

It introduces a novel autoencoder-based method combined with differential evolution algorithms for classifying complex correlation functions from protein phase separation experiments.

Findings

Successfully classifies experimental correlation maps by quench depth and concentration.

Automates analysis of large dynamic X-ray scattering datasets.

Facilitates rapid comparison between experimental data and phase separation models.

Abstract

We use machine learning methods for an automated classification of experimental XPCS two-time correlation functions from an arrested liquid-liquid phase separation of a protein solution. We couple simulations based on the Cahn-Hilliard equation with a glass transition scenario and classify the measured correlation maps automatically according to quench depth and critical concentration at a glass/gel transition. We introduce routines and methodologies using an autoencoder network and a differential evolution based algorithm for classification of the measured correlation functions. The here presented method is a first step towards handling large amounts of dynamic data measured at high brilliance synchrotron and X-ray free-electron laser sources facilitating fast comparison to phase field models of phase separation.