Massively parallel unsupervised single-particle cryo-EM data clustering via statistical manifold learning

Abstract

Structural heterogeneity in single-particle cryo-electron microscopy (cryo-EM) data represents a major challenge for high-resolution structure determination. Unsupervised classification may serve as the first step in the assessment of structural heterogeneity. However, traditional algorithms for unsupervised classification, such as K-means clustering and maximum likelihood optimization, may classify images into wrong classes with decreasing signal-to-noise-ratio (SNR) in the image data, yet demand increased computational costs. Overcoming these limitations requires further development of clustering algorithms for high-performance cryo-EM data processing. Here we introduce an unsupervised single-particle clustering algorithm derived from a statistical manifold learning framework called generative topographic mapping (GTM). We show that unsupervised GTM clustering improves classification accuracy by about 40% in the absence of input references for data with lower SNRs. Applications to several experimental datasets suggest that our algorithm can detect subtle structural differences among classes via a hierarchical clustering strategy. After code optimization over a high-performance computing (HPC) environment, our software implementation was able to generate thousands of reference-free class averages within hours in a massively parallel fashion, which allows a significant improvement on ab initio 3D reconstruction and assists in the computational purification of homogeneous datasets for high-resolution visualization.

Conformational Landscape of the p28-Bound Human Proteasome Regulatory Particle

“…42 classes were selected for the generation of an initial model using e2initialmodel.py. In the following steps, reference-free 2D classification was performed in ROME (Wu et al, 2016), whereas unsupervised 3D classification was performed in RELION 1.3 (Scheres, 2012b). After the first round of reference-free 2D classification using all 250,251 particles binned to a pixel size of 4.00 Å, bad particles and non-RP class averages were rejected as a whole class upon inspection of class average quality.…”

Chemistry of cation hydration and conduction in a skeletal muscle ryanodine receptor

“…After auto-picked and verified using RELION 2.1 (41) and EMAN2 software (42), 835,709 particles of RyR1 were extracted from 9,528 good micrographs for the following analysis. All reference-free 2D and 3D classifications were carried out in RELION 2.1/3.0 (41) and ROME 1.1 (43), which combined the regularized maximum-likelihood based image alignment and manifold learning-based classification. 3D refinement that refined the Euler angles and x/y-shifts of each particle to further improve the resolution of density maps was completed in RELION.…”

“…After several rounds of iterative 2D and 3D classifications, bad particles and non-particle In order to improve the local resolution of density maps, the CTF parameters of each individual particle were refined locally with program Gctf (40). Based on the x/y-shift and orientation parameters of each particle from the high-resolution 3D refinement, we reconstructed two halfmaps of each state using raw single-particle images at super-resolution mode with a pixel size of 0.685 Å by ROME software (43). The global resolutions of the finial reconstructions of states 1, 2 and 3 were 2.6, 3.3 and 6.3 Å respectively, measured by the gold-standard Fourier shell correlation (FSC) at 0.143-cufoff on two independently refined half-maps.…”

DNA Origami as Scaffolds for Self‐Assembly of Lipids and Proteins

“…[45] Single-particlec ryo-EM has been recentlye mployed to determine both protein structures and dynamics in their native, or even intermediate functional states, down to nearatomic resolution, [46] thanks to the application of direct electron detectors, as well as that of machine-learning-based data processing technology. [47] The direct electron detector has significantly enhanced the signal-to-noise ratio of recordedc ryoelectron micrographs.T he entire structures of DNA origami assembliesc an be preferably analyzed in 3D through single-par-ticle cryo-EM analysis. In 2012, Bai et al reported the first DNA origami structure resolved by using cryo-EM, with an overall resolution of 11.5 .…”