SPEAKER: Dr. Alfredo Bellisario (Uppsala University)
TITLE: Structure from Explosions: Deep Learning on Coulomb Explosions and Single-Particle Imaging with Orientation Bias
Abstract:
Single-particle imaging (SPI) with X-ray free-electron lasers (XFELs) offers a powerful method for studying the three-dimensional structure of individual biomolecules in the gas phase.
However, the lack of control over particle orientation—due to aerosol-based sample delivery— limits the efficiency and resolution of the reconstructions. Relative particle orientations are typically determined using algorithms such as Expectation-Maximization Compression (EMC).
Recent experimental developments have introduced techniques for pre-aligning particles using electric fields, induced by optical-laser pumping, before they interact with the X-ray beam. This orientation biasing provides prior information that can be exploited to reduce the number of patterns required for achieving highresolution 3D reconstructions (Wollter 2024).
In this context, Coulomb explosion imaging has emerged as a promising complementary technique. By tracing the fragments and momenta of the resulting ions in coincidence, Coulomb explosion imaging enables structural determination of a sample stripped of its electrons. Recent advances have demonstrated that measuring ions from laser-induced Coulomb explosions can reveal structural features (André 2024), even with partial detector coverage or without coincidence detection. These ion signatures can provide partial information about particle orientation, which can be integrated into existing reconstruction algorithms such as EMC to improve efficiency and reduce the reliance on high-quality diffraction patterns (André 2025). Moreover, these explosion signatures can be correlated with protein structure, potentially enabling structural determination directly from explosion data. Coupled with machine learning, this approach could potentially allow molecular structure prediction without relying on diffraction.
