Unraveling protein crystallization in ionic liquids via high-throughput SAXS and data-driven machine learning

Ionic liquids (ILs) can modulate protein phase behavior, including crystallization and aggregation. However, crystallization outcomes are often difficult to predict because nucleation is stochastic and strongly affected by specific ion effects. Here, we examine the crystallization pathway of lysozyme in ethylammonium nitrate (EAN) using a multi-modal small-angle X-ray scattering (SAXS) strategy that integrates high-throughput 96-well screening, in situ capillary thermal treatment, and time-resolved kinetic monitoring. To address batch-to-batch variability in these soft-matter samples, we implemented a standardized workflow for extracting complementary SAXS descriptors, including a fixed-window pseudo-R_g descriptor of low-q scattering evolution and a crystallinity index based on Bragg-feature intensity. The SAXS data are consistent with a two-step nucleation-like pathway in which EAN-induced association precedes the appearance of long-range crystalline order, while thermal treatment disrupts metastable clusters without restoring a simple native monomeric state. Machine-learning models were used as within-system interpolation and screening-prioritization tools for the measured lysozyme-EAN composition grid. Linear models failed to reproduce the non-linear trends, whereas ensemble methods reproduced the main features of the measured crystallization region. Feature-importance analysis identified EAN concentration as the dominant control variable (approximately 70% contribution). Together, these results provide a SAXS-based descriptor-extraction and data-analysis workflow for mapping composition-dependent crystallization behavior in the lysozyme-EAN system.