Smartscope: AI-driven grid navigation for high-throughput cryo-EM
Specimen optimization is currently one of the main limiting steps in the cryo-electron microscopy (EM) structure determination pipeline. The ideal specimen is a molecule-thin layer of macromolecules in solution frozen on top of a holey membrane stabilized by a metal support grid. During screening, experienced microscopists visualize the specimen at increasing magnifications by navigating to areas that are most likely to provide information useful to guide the optimization. Iterating this procedure over different experimental conditions, eventually results in grids that are suitable for high-resolution imaging. While automation has led to increased throughput of data collection in single particle cryo-EM, specimen screening is still a largely manual and time-consuming task where data coherence and intermediate readouts are not frequently recorded. Here, we present Smartscope, a framework to simplify and automate the screening process of cryo-EM grids. By abstracting the intermediate steps of specimen navigation, Smartscope saves metadata into a database and presents the results to the user through an interactive, user-friendly web based interface. Grid squares and holes in the substrate are automatically detected and labeled using neural network-based approaches that si-multaneously detect and classify squares with high accuracy, and precisely recover the position of holes within grid squares. Moreover, Smartscope's web interface can also be used as a platform for automated data collection as it allows the quick selection of areas for imaging, thus significantly reducing setup time. By unifying data management for proper bookkeeping and using AI-based routines for autonomous grid navigation, Smartscope offers a convenient platform that minimizes human intervention and optimizes microscope usage, thus significantly improving the throughput of cryo-EM structure determination.
