Figure Open Access

# Visualization of SARS-CoV-2 Infection Scenes by "Zero-Shot" Enhancements of Electron Microscopy Images

Drefs, Jakob; Salwig, Sebastian; Lücke, Jörg

"Zero-Shot" enhancements of an electron microscopy image of SARS-CoV-2  viruses in Vero cell cultures using probabilistic machine learning algorithms for denoising.  The data available here were obtained and are discussed in the paper Visualization of SARS-CoV-2 Infection Scenes by "Zero-Shot" Enhancements of Electron Microscopy Images by Drefs et al. (2021). As input we used data made available by Laue et al. (2021) who recorded images of ultrathin plastic sections using transmission electron microscopy (we downloaded the data from this Zenodo repository). The input image can be found in the H5 file sars-cov2-em-noisy-input.h5. Based on the data, we estimated pixel means and variances during the application of probabilistic machine learning algorithms for denoising. In the H5 files sars-cov2-em-sssc-mean-reconstruction.h5 and sars-cov2-em-sssc-variance-reconstruction.h5 the mean and variance of pixel estimations obtained with a Spike-and-Slab Sparse Coding (SSSC) model can be found (illustrated in Fig. 2 in the paper by Drefs et al. (2021)). In the H5 files sars-cov2-em-gpmm-mean-reconstruction.h5 and sars-cov2-em-gpmm-variance-reconstruction.h5 the mean and variance of pixel estimations obtained with a Gamma Poisson Mixture model (GPMM) can be found (illustrated in Fig. 3 in the paper by Drefs et al. (2021)). The image "sars-cov2-em-sssc-variance-reconstruction-colorized.png" (illustrated in Fig.1 in the paper by Drefs et al. (2021)) was obtained after contrast enhancement and colorization: structures that we manually identified as belonging to a cell were colored in blue, the remainder was colorized in yellow.

The H5 files can be read and visualized in Python as follows:

import glob
import h5py
import matplotlib.pyplot as plt
for file in glob.glob("*.h5"):
with h5py.File(file, "r") as f:
plt.figure()
plt.imshow(f["data"][...], cmap="gray")
plt.title(file)
plt.show()

We would like to acknowledge funding by the German Ministry of Research and Education (BMBF) in the project 05M2020 (SPAplus) which enabled this research through a top-up fund for COVID-19 research; and we would like to acknowledge funding by the DFG project 352015383 (SFB 1330, B2) which provided source code to train generative models. Furthermore, we would like to acknowledge support in terms of computational resources by the Oldenburg High Performance Compute Cluster (CARL) and by the North German Supercomputing Alliance under grant nim00006.
Files (10.4 MB)
Name Size
sars-cov2-em-gpmm-mean-reconstruction.h5
md5:88feeb70a50b5156c37f923135a5edb3
2.4 MB
sars-cov2-em-gpmm-variance-reconstruction.h5
md5:d49489eccd04f2f59c87d8b72bd72377
2.4 MB
sars-cov2-em-noisy-input.h5
md5:9acd027f1f66c016fcd2efbba37d3b72
526.3 kB
sars-cov2-em-sssc-mean-reconstruction.h5
md5:237cde6de13ae1f1728aca1a11a9d9dd
2.4 MB
sars-cov2-em-sssc-variance-reconstruction-colorized.png
md5:8224f1878e3212262baac10b3554bc37
392.7 kB
sars-cov2-em-sssc-variance-reconstruction.h5
md5:408e3d997fc2b34bd4957ac047fe328d
2.4 MB
296
27
views