A comprehensive foundation model for cryo-EM image processing

Abstract Cryogenic electron microscopy (cryo-EM) has become a premier technique for high-resolution structural determination of biological macromolecules 1–4 . However, its widespread adoption is hampered by the need for specialized expertise. We introduce the Cryo-EM Image Evaluation Foundation (Cryo-IEF) model, pre-trained on an extensive dataset of approximately 65 million cryo-EM particle images using unsupervised learning. Cryo-IEF excels in various cryo-EM data processing tasks, such as classifying particles from different structures, clustering particles by pose, and assessing the quality of particle images. Upon fine-tuning, the model effectively ranks particle images by quality at high efficiency, enabling the creation of CryoWizard—a fully automated single-particle cryo-EM data processing pipeline. This pipeline has successfully resolved high-resolution structures of diverse properties and proven adept at mitigating the prevalent preferred orientation challenge in many cryo-EM samples. The Cryo-IEF model and CryoWizard pipeline collectively represent a significant advancement in rendering cryo-EM technology more accessible, efficient, and robust, with substantial implications for life sciences research.