This project aims to develop machine-learning methods for plasma simulations and molecular fragmentation for X-ray-free electron lasers (XFELs). XFELs enabled new imaging techniques that are very interesting for studying biological molecules and dynamical processes. In particular, these powerful X-ray light sources produce ultra-fast, coherent X-ray pulses with peak brilliance billions of times higher than synchrotrons. The "diffraction before destruction'' principle ensures the elastically scattered photons collected by the detectors provide structural information within the very few-femtoseconds of photon-matter interaction before radiation damage occurs. Yet, in the high-intensity radiation regime of XFELs, the sample turns into plasma due to the extreme radiation dose. The integration of spectroscopical data, fluorescence, and the time-of-flight of exploding ions provides the full picture of the several dynamical processes involved in these complex experiments. Studying dynamical variables of plasma such as electron temperature, ion temperature, and ionization allows an understanding of the structural changes in the biological sample, in particular, they allow calculation of structure factors, cross-section, and ultimately the estimate of the understanding of experimental conditions. From a computational perspective, the theoretical analysis of all these processes is extremely challenging to tackle. Within this project, we develop ML methods to enhance our simulations and the understanding of these phenomena. We apply our results to support the method development of experiments that aim for example to provide a deeper understanding of molecular structures and dynamics in proteins. To address the aforementioned challenges, we are developing ML computational to: - Characterize and simulate X-ray emission spectra from protein samples. We will analyze synthetic spectra generated using plasma simulation codes. Simulating plasma processes currently requires many hours on multiple CPU cores. ML can help optimize these simulations. - Predict X-ray pulse parameters and experimental conditions (i.e. ion temperature) from experimental data. By leveraging ML models trained on experimental and simulated data, we aim to extract key plasma properties that influence molecular dynamics and radiation damage. - Gain structural knowledge from molecular fragmentation ion maps and diffraction. We employ molecular fragmentation simulations, including those performed with MolDStruct [2] and other in-house developed codes, to generate synthetic ion maps. These maps help address critical challenges, such as determining sample orientation in single-particle imaging diffraction experiments [3] and understanding fragmentation pathways. These results are very recent and we are just now exploring the possibilities of ML in this field. References: [1.] Ekeberg, T., et. al. (2024). Observation of a single protein by ultrafast X-ray diffraction. Light: Science & Applications, 13(1), 15. [2.] Dawod, I., et. al. (2024); MolDStruct: Modeling the dynamics and structure of matter exposed to ultrafast x-ray lasers with hybrid collisional-radiative/molecular dynamics. J. Chem. Phys. 14 May 2024; 160 (18): 184112 [3.] André, T., De Santis, E., Timneanu, N., & Caleman, C. (2025). Partial Orientation Retrieval of Proteins From Coulomb Explosions. Accepted in Physical Review Letters.