RE2DC: a robust and efficient 2D classifier with visualization for processing massive and heterogeneous cryo-EM data

Abstract

Single-particle cryo-electron microscopy (cryo-EM) increasingly generates millions of particle images, yet two-dimensional (2D) classification remains a major bottleneck because existing approaches balance computational efficiency against robustness to noise, outliers and structural heterogeneity. We introduce RE2DC (Robust and Efficient 2D Classifier), an algorithmic framework that resolves this trade-off through dynamic linear-time clustering, dimension-reduction multi-reference alignment, and offers real-time interactive t-SNE visualization. Rather than relying primarily on hardware acceleration, RE2DC reduces the computational cost of robust clustering and employs de-noised images for alignment, enabling efficient execution on standard multi-core CPUs. Across diverse benchmark datasets, RE2DC achieves class homogeneity comparable to ISAC while processing datasets three- to ten-fold faster per classification round than RELION. Notably, RE2DC resolves rare, structurally coherent particle populations, enabling detection of transient conformational intermediates and supporting near real-time cryo-EM analysis. By addressing algorithmic complexity, RE2DC establishes a general framework for robust, scalable analysis of massive and heterogeneous image datasets.