WP1: Machine Learning to optimize performance of advanced synchrotron and FEL light-sources

Work Package 1 dealt with the application of machine learning algorithms to optimize accelerator-based X-ray facilities.

The main work of this work package can be grouped into three tasks:

“Our work in the project was to create some strategy or hierarchy for the sheer amount of tuning required in these machines.”

Read the Q&A with work package co-lead Annika Eichler on how the work package used machines to improve tuning and reduce noise from environmental factors like seismic waves

“What surprised me… is that there is no ‘one-button solution’ for data analysis & diagnostics.”

Read the Q&A with early-career researcher Gesa Goetzke on her machine learning research during the project

Key results include:

Testing how Large Language Models such as GPT 4 and Llama 2 can work with humans to tune a particle accelerator

Kaiser, Jan, Anne Lauscher, and Annika Eichler. "Large language models for human-machine collaborative particle accelerator tuning through natural language." Science advances 11.1 (2025): eadr4173.

Using AI and old logbooks to determine how to best tune FELs

Sulc, Antonin, et al. "Towards unlocking insights from logbooks using ai." arXiv preprint arXiv:2406.12881 (2024).

WP2: High-throughput data handling and vetoing

This work package developed strategies to better cope with the extremely high data rates caused by advanced X-ray imagers and X-ray sources with extremely high average brightness. Novel concepts were developed and tested for the case of Serial Femtosecond Crystallography (SFX), where the impact of improved data handling is potentially very large.

“Handling the large amounts of data produced by this generations of experiments is a big task. One step in that task is to do data reduction as early as possible to reduce the workload for later stages of processing.”

Read the Q&A with WP co-lead David Pennicard on how the work package developed X-ray detectors and quick data reduction methods to filter out ‘bad data’

Key results include:

An improved image analysis method called MP-FAST that can identify useful image data by using hardware like CPUs, GPUs, and FPGAs to speed things up

Rahmani, Vahid, et al. "Robust image descriptor for machine learning based data reduction in serial crystallography." Applied Crystallography 57.2 (2024): 413-430.

Creating a real-time speckle pattern classification model called SpeckleNN for X-ray single-particle imaging with limited labeled examples

Wang, Cong, et al. "SpeckleNN: a unified embedding for real-time speckle pattern classification in X-ray single-particle imaging with limited labeled examples." IUCrJ 10.5 (2023): 568-578.

WP3: Automated and robotic sample delivery for high-throughput experiments

This work package developed reliable, robust, high throughput sample delivery techniques that are crucial to optimally use the facilities at the very high repetition rate, free-electron laser sources such as FLASH and European XFEL in Hamburg and LCLS-II in Stanford for breakthrough science.

“One of my major goals has been to make measurement techniques more efficient and automated.”

Read the Q&A with WP co-lead Alke Meents on new fixed target delivery systems for FELs, and the importance of automation, including the use of robots for sample exchange and data analysis

Key results:

Finding the active site and inhibitors in the SARS-CoV-2 main protease

Günther, Sebastian, et al. "X-ray screening identifies active site and allosteric inhibitors of SARS-CoV-2 main protease." Science 372.6542 (2021): 642-646.

Characterising the antiviral activity of Covalent CathepsinL Inhibitors

Falke, Sven, et al. "Structural elucidation and antiviral activity of covalent cathepsin L inhibitors." Journal of medicinal chemistry 67.9 (2024): 7048-7067.

WP4: High-performance X-ray optics

This work package focused on providing and maintaining the highest quality X-ray optics for the scientific use of high-brightness X-ray sources. The participating labs jointly worked on the development of efficient cooling schemes, strategies to find optimum mirror coatings for long-term durability and performance, and mitigation and possible in-situ removal of surface contaminants.

“Overall we both understood more about the damage threshold of these mirror linings, and also how to run better simulations.”

Read the Q&As with WP co-leaders Maurizio Vannoni and Elke Plönjes-Palm on the contamination issues with X-ray optics, particularly for high-intensity lasers, and the importance of mirror coatings to improve durability

Key results:

Determining whether the optical delay line can handle the high peak power of a single X-ray pulse and explores ways to prevent damage

Tavakkoly, Marziyeh, et al. "Simulations about stability, damage, and heating impacts for an x-ray optical delay line at sase3." Journal of Physics: Conference Series. Vol. 2380. No. 1. IOP Publishing, 2022.

Developing computer simulations of how X-rays travel to the sample and experimental results showing two-color X-ray beams at different energy levels

Serkez, Svitozar, et al. "Opportunities for two-color experiments in the soft X-ray regime at the European XFEL." Applied Sciences 10.8 (2020): 2728.