An unattended image-processing pipeline for on-the-fly quality assessment and 3D exploration in cryo-EM

Single-particle analysis (SPA) by cryogenic electron microscopy (cryo-EM) has become a cornerstone of structural biology; yet, the workflow from data acquisition to a 3D structure remains a time-consuming, manual and significant task. Researchers often invest days of valuable microscope time collecting massive datasets with little to no real-time feedback on sample quality or the ultimate feasibility of achieving a high-resolution reconstruction. To address this challenge, we have developed an automated, end-to-end processing workflow designed for on-the-fly analysis. Built within the Scipion framework, the pipeline integrates a cascade of automated quality-control filters, a novel consensus-based strategy for training a data-specific particle-picking model, and a parallel 2D/3D validation scheme to ensure robust processing outcomes. We validated the workflow on a diverse benchmark of 32 datasets (CryoPPP), achieving a 94% overall processing success rate and obtaining high-quality 3D reconstructions in 78% of cases. Real-world deployment at the European Synchrotron (ESRF) cryo-EM facility further validated its practical utility: the pipeline demonstrated processing speeds exceeding data acquisition, delivering preliminary 3D maps in under 3 h. Approximately 70% of user experiments converged to interpretable structures, with half achieving 3-4 Å resolution, despite the inherent complexity of facility-collected samples. By providing researchers with rapid, actionable feedback and identifying both promising and problematic datasets in real time, this workflow transforms cryo-EM data collection from a passive process into an active, data-driven experiment. It serves as a powerful diagnostic and decision-support tool, accelerating structural determination and optimizing microscope time in high-throughput cryo-EM environments.