Leopard-EM: an extensible 2D template-matching package to accelerate in situ structural biology.

The ability to generate high-resolution views of cells with cryogenic electron microscopy (cryo-EM) can reveal the molecular mechanisms of biological processes in their native cellular context. The revolutionary impact of this strategy is limited by the difficulty of accurately annotating structures within these images. 2D template matching (2DTM), in which high-resolution structural models are used as computational probes to locate and orient molecular complexes with high precision, has shown initial promise in annotating single molecules in cellular cryo-EM images. While the scientific community works to identify best practices for applying 2DTM to specific biological questions and to maximize sensitivity and throughput, a modular and extensible software architecture would support the rapid development of novel methodological approaches, thus accelerating innovation within the field. To achieve this, we developed Leopard-EM (Location and orientation of particles found using two-dimensional tEmplate Matching), a modular Python-based 2DTM implementation built to be easily customizable. We implemented an automated pixel-size refinement procedure and find that 2DTM is sensitive to pixel-size errors of ∼0.2%. To demonstrate the flexibility of the Leopard-EM architecture, we developed a constrained search protocol that improved small ribosomal subunit detection by approximately eightfold by using initial locations and orientations determined for the large ribosomal subunit. Using this strategy, we captured a distribution of ribosome rotation states within the living cell at single-molecule resolution. We envision that Leopard-EM can be used as a platform for the development of in situ cryo-EM data-processing workflows, facilitating the rapid development of this field. Leopard-EM is available at https://github.com/Lucaslab-Berkeley/Leopard-EM.