REST: A method for restoring signals and revealing individual macromolecule states in cryo-ET

Abstract

Cryo-electron tomography (cryo-ET) is widely used to explore the 3D density of biomacromolecules. However, the heavy noise and missing wedge effect prevent directly visualizing and analyzing the 3D reconstructions. Here, we introduced REST, a deep learning strategy-based method to establish the relationship between low-quality and high-quality density and transfer this knowledge to restore signals in cryo-ET. Experimental results on purified ribosome and recombinant nucleosome datasets showed that REST had outstanding performance in denoising and compensating the missing wedge. The application in dynamic nucleosome structures suggests that REST has the capability to reveal individual macromolecules which present different conformations without subtomogram averaging. Moreover, REST could greatly improve the reliability of particle picking. These advantages enable REST to be a powerful tool for the straightforward interpretation of target macromolecules by visual inspection of the density and of a broad range of other applications in cryo-ET, such as segmentation, particle picking, and subtomogram averaging.

An image processing pipeline for electron cryo-tomography in RELION-5

“…Gaining structural insights from cryo-ET data is complex and remains an active area of research, with ongoing developments in CTF estimation 39 , tilt series alignment 8 , tomogram denoising 21,23,24,45,52 , automated localisation of structures of interest 30,[53][54][55][56][57][58][59][60][61][62] and approaches to deal with structural heterogeneity among extracted particles [63][64][65][66] , amongst many others. As a result, cryo-ET image processing pipelines, including the one described here, will need to be continuously updated in the coming years.…”

“…Direct reconstruction of local regions of interest in tomograms 1,[17][18][19][20] has obviated the need for calculating large, high-resolution tomograms from which subvolumes are cropped. To assist in feature identification, deep-learning approaches have gained traction for denoising [21][22][23][24] and template matching 25,26 .…”

Nanoscale analysis of human G1 and metaphase chromatinin situ

“…Nucleosome linker DNA orientation and length variability may both contribute to the nucleosome class averages that have short linker-DNA densities. Previous work on isolated chromosomes and oligonucleosomes in vitro produced tomograms of unambiguous nucleosomes (Beel et al , 2021; Zhang et al , 2023; Zhang et al , 2022). While it is possible to directly visualize DNA in purified chromatin, which contains nucleosome and DNA densities, such a task is not feasible in situ , which contains many non-chromatin densities.…”

DUAL: deep unsupervised simultaneous simulation and denoising for cryo-electron tomography

“…DUAL achieves unpaired/unsupervised training by adversarial learning. In paired image-to-image translation training, for each image, the ground truth translation target image is required for learning their correspondence relationship [52]. In comparison, unpaired image inputs from the two domains are sufficient in the unpaired/unsupervised setting.…”

“…Meanwhile, deep learning based methods avoid modeling the noise pattern explicitly. Supervised approach requires carefully prepared ground truth denoised version of tomograms by averaging and aligning structures [52]. Unsupervised approaches have been proposed to learn from 2D projection images.…”