CASSPER: A Semantic Segmentation based Particle Picking Algorithm for Single Particle Cryo-Electron Microscopy

Abstract

Single-particle cryo-electron microscopy has emerged as the method of choice for structure determination of proteins and protein complexes. However, particle identification and selection which is a prerequisite for achieving high-resolution still poses a major bottleneck for automating the steps of structure determination. Here, we present a generalised deep learning tool, CASSPER, for the automated detection and isolation of protein particles in transmission microscope images.This deep learning tool uses Semantic Segmentation and a collection of visually prepared training samples to capture the differences in the transmission intensities of protein, ice, carbon and other impurities found in the micrograph. CASSPER is the first method to do pixel level classification and completely eliminates the need of manual particle picking. Integration of Contrast Limited Adaptive Histogram Equalization (CLAHE) in CASSPER enables high-fidelity particle detection even in micrographs with variable ice thickness and contrast. In addition, our generalized model for cross molecule picking works with high efficiency on unseen datasets and can potentially pick particles on-the-fly, thereby, enabling automation of data processing. This work in J. Kumar lab is supported by the Wellcome Trust DBT India Alliance fellowship (grant number IA/I/13/2/501023). B. George thanks University Grants Commission and CMS College Kottayam for the Teacher Fellowship under FDP Scheme. A. Assaiya thanks DBT, India for a Senior Research Fellowship. We are also thankful to Dr. K. Vaghmare, and computational facility at IUCAA, Pune for helping with transfer and storage of raw EM data. Author ContributionsJK, RC, AKK, NSP and GP formulated the project and initiated the discussions at NCCS and IUCAA. The preprocessing and deep learning tools were developed and implemented by NSP, BG and RJR at airis4D. BG and RJR did testing and troubleshooting with inputs from AK, RC, AA and JK. The uniform pipeline, high-resolution 3D reconstruction and comparison with other tools were done by AA at NCCS with inputs from RC, NSP, AK and JK. The manuscript was read and approved by all the authors before submission.

CryoPPP: A Large Expert-Labelled Cryo-EM Image Dataset for Machine Learning Protein Particle Picking

“…A common problem of the particle picking algorithms trained on a small amount of particle data of a few proteins is that they cannot distinguish 'good' and 'bad' particles well, including overlapped particles, local aggregates, ice contamination and carbon-rich areas 31 . For instance, the methods: DRPnet 32 , TransPicker 33 , CASSPER 34 , and McSweeney et al's method 35 that made significant contributions to the particle selection problem suffered the two similar problems. Firstly, there is not a sufficient and diversified dataset to train them.…”

Review: Deep Learning in Electron Microscopy

“…Labelling ANNs are often combined with other methods. For example, ANNs can be used to automatically identify particle locations 185,[284][285][286] to ease subsequent processing.…”

“…To improve performance, DNNs have been trained to semantically segment images [301][302][303][304][305][306][307][308] . Semantic segmentation DNNs have been developed for focused ion beam scanning electron microscopy [309][310][311] (FIB-SEM), SEM [311][312][313][314] , STEM 287,315 , and TEM 286,310,311,[316][317][318] . For example, applications of a DNN to semantic segmentation of STEM images of steel are shown in figure 3.…”

Disentangling multiple scattering with deep learning: application to strain mapping from electron diffraction patterns

“…Artificial intelligence and machine learning (AI/ML) algorithms are increasingly being implemented in materials characterization, including in electron microscopy 42 . Deep-learning approaches have been been demonstrated to outperform classical algorithms in variety of computer vision problems in microscopy including classification and segmentation problems [43][44][45] . For instance, deepconvolutional neural networks (CNNs) are implemented in the analysis of images collected with various microscopy techniques such as crystal phase classification from back-scattered diffraction patterns 46 , structure measurement from electron diffraction and atomic-resolution STEM images 47 and from scanning tunneling microscopy 48 , crystal symmetry identification from X-ray diffraction 49 , defect analysis from atomic-resolution STEM images 50 , crystal tilt and thickness detection from position averaged CBED patterns 51,52 , and orientation and strain mapping from 4D-STEM diffraction datasets 53,54 .…”