Enhancing SNR and generating contrast for cryo-EM images with convolutional neural networks

In cryogenic electron microscopy (cryo-EM) of radiation-sensitive biological samples, both the signal-to-noise ratio (SNR) and the contrast of images are critically important in the image processing pipeline. Classic methods improve low-frequency image contrast experimentally, by imaging with high defocus, or computationally, by applying various types of low-pass filter. These contrast improvements typically come at the expense of high-frequency SNR, which is suppressed by high-defocus imaging and removed by low pass filtration. Here, we demonstrate that a convolutional neural network (CNN) denoising algorithm can be used to significantly enhance SNR and generate contrast in cryo-EM images. We provide a quantitative evaluation of bias introduced by the denoising procedure and its influences on image processing and three-dimensional reconstructions. Our study suggests that besides enhancing the visual contrast of cryo-EM images, the enhanced SNR of denoised images may facilitate better outcomes in the other parts of the image processing pipeline, such as classification and 3D alignment. Overall, our results provide a ground of using denoising CNNs in the cryo-EM image processing pipeline.

[1]  Tim Salimans,et al.  Weight Normalization: A Simple Reparameterization to Accelerate Training of Deep Neural Networks , 2016, NIPS.

[2]  G J Jensen,et al.  Alignment error envelopes for single particle analysis. , 2001, Journal of structural biology.

[3]  Neil J. Bershad,et al.  On estimating signal-to-noise ratio using the sample correlation coefficient (Corresp.) , 1974, IEEE Trans. Inf. Theory.

[4]  Alan Brown,et al.  Author response: Cryo-EM structure of the Plasmodium falciparum 80S ribosome bound to the anti-protozoan drug emetine , 2014 .

[5]  J. Frank,et al.  Signal-to-noise ratio of electron micrographs obtained by cross correlation , 1975, Nature.

[6]  David J. Fleet,et al.  cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination , 2017, Nature Methods.

[7]  Yuan Yu,et al.  TensorFlow: A system for large-scale machine learning , 2016, OSDI.

[8]  D. Agard,et al.  MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy , 2017, Nature Methods.

[9]  Wolfgang Baumeister,et al.  Using the Volta phase plate with defocus for cryo-EM single particle analysis , 2016, bioRxiv.

[10]  Wolfgang Baumeister,et al.  Author response: Using the Volta phase plate with defocus for cryo-EM single particle analysis , 2017 .

[11]  S. Scheres,et al.  Cryo-EM structure of the Plasmodium falciparum 80S ribosome bound to the anti-protozoan drug emetine , 2014, eLife.

[12]  K. Nagayama,et al.  Transmission electron microscopy with Zernike phase plate. , 2001, Ultramicroscopy.

[13]  Erik Lindahl,et al.  New tools for automated high-resolution cryo-EM structure determination in RELION-3 , 2018, eLife.

[14]  Matthew L. Baker,et al.  A 9 Å single particle reconstruction from CCD captured images on a 200 kV electron cryomicroscope , 2004 .

[15]  Bonnie Berger,et al.  Positive-unlabeled convolutional neural networks for particle picking in cryo-electron micrographs , 2018, Nature Methods.

[16]  Gabriel C. Lander,et al.  High-resolution structure determination of sub-100 kDa complexes using conventional cryo-EM , 2019, Nature Communications.

[17]  Robert M Glaeser,et al.  Laser phase plate for transmission electron microscopy , 2019, Nature Methods.

[18]  Thomas Brox,et al.  U-Net: Convolutional Networks for Biomedical Image Segmentation , 2015, MICCAI.

[19]  N. Grigorieff,et al.  Accurate determination of local defocus and specimen tilt in electron microscopy. , 2003, Journal of structural biology.

[20]  Loïc Royer,et al.  Noise2Self: Blind Denoising by Self-Supervision , 2019, ICML.

[21]  R. Glaeser.  Review: electron crystallography: present excitement, a nod to the past, anticipating the future. , 1999, Journal of structural biology.

[22]  Robert M. Glaeser,et al.  Laser-Based Zernike Plate for Phase Contrast Transmission Electron Microscopy , 2018 .

[23]  Nikolaus Grigorieff,et al.  Measuring the optimal exposure for single particle cryo-EM using a 2.6 Å reconstruction of rotavirus VP6 , 2015, eLife.

[24]  Guigang Zhang,et al.  Deep Learning , 2016, Int. J. Semantic Comput..

[25]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[26]  Ning Xu,et al.  Wide Activation for Efficient and Accurate Image Super-Resolution , 2018, ArXiv.

[27]  Lei Zhang,et al.  Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising , 2016, IEEE Transactions on Image Processing.

[28]  Kai Zhang,et al.  Gctf: Real-time CTF determination and correction , 2015, bioRxiv.

[29]  Sjors H.W. Scheres,et al.  Faculty Opinions recommendation of Real-time cryo-electron microscopy data preprocessing with Warp. , 2019, Faculty Opinions – Post-Publication Peer Review of the Biomedical Literature.

[30]  Erik Lindahl,et al.  New tools for automated high-resolution cryo-EM structure determination in RELION-3 , 2018, eLife.

[31]  Vincent Dumoulin,et al.  Deconvolution and Checkerboard Artifacts , 2016 .

[32]  T. Grant,et al.  Single particle cryo-EM structure of rotavirus VP6 at 2.6 Angstrom resolution , 2015 .

[33]  Tristan Bepler,et al.  Topaz-Denoise: general deep denoising models for cryoEM and cryoET , 2019, Nature Communications.

[34]  Wen Jiang,et al.  A 9 angstroms single particle reconstruction from CCD captured images on a 200 kV electron cryomicroscope. , 2004, Journal of structural biology.

[35]  R. H. Wade.  A brief look at imaging and contrast transfer , 1992 .

[36]  J. Frank,et al.  Determination of signal-to-noise ratios and spectral SNRs in cryo-EM low-dose imaging of molecules. , 2009, Journal of structural biology.

[37]  Daniel Asarnow,et al.  Structure of the human TRPM4 ion channel in a lipid nanodisc , 2018, Science.

[38]  Michael Habeck,et al.  Bayesian inference of initial models in cryo-electron microscopy using pseudo-atoms. , 2015, Biophysical journal.

[39]  Jaakko Lehtinen,et al.  Noise2Noise: Learning Image Restoration without Clean Data , 2018, ICML.

[40]  Yifan Cheng.  Single-Particle Cryo-EM at Crystallographic Resolution , 2015, Cell.

[41]  Geoffrey E. Hinton,et al.  ImageNet classification with deep convolutional neural networks , 2012, Commun. ACM.

[42]  Saburo Ito,et al.  Cryo-EM Reveals Integrin-Mediated TGF-β Activation without Release from Latent TGF-β , 2020, Cell.