DeepBacs – Escherichia coli nucleoid denoising dataset and CARE model

Training and test images of H-NS-mScarlet-I expressing E. coli cells for image denoising, as well as a trained CARE model.

Additional information can be found on our github wiki.

The example images show confocal images of labelled E. coli nucleoids at low and high SNR.

Training and test dataset

Data type: Paired microscopy images (fluorescence)

Microscopy data type: Confocal fluorescence images

Microscope: Leica SP8 confocal microscope with a 1.40 NA 63x oil immersion objective 

Cell type: E. coli strain CS01 expressing H-NS-mScarlet-I fusion protein (H-NS-mScarlet-I) in NO34 parental strain (MreB-sfGFPsw, kindly provided by Zemer Gitai) 

File format: .tif (16-bit)

Image size: 512 x 512 px2 (Pixel size: 45 nm)

CARE model:

The CARE 2D model was generated using the ZeroCostDL4Mic platform (Chamier et al., 2021). It was trained from scratch for 100 epochs (600 steps/epoch) on 1400 paired image patches (image dimensions: (512 x 512 px²), patch size: (64 x 64 px²), 50 patches/image) with a batch size of 8 and a laplace loss function, using the CARE 2D ZeroCostDL4Mic notebook (v 1). Key python packages used include tensorflow (v 0.1.12), Keras (v2.3.1), csbdeep (v 0.6.2), numpy (v 1.19.5), cuda (v 11.0.221). The training was accelerated using a Tesla T4 GPU and data was augmented by a factor of 4 using rotation and flipping.

The model weights can be used with the ZeroCostDL4Mic CARE 2D notebook and the CSBDeep Fiji plugin.

Author(s): Christoph Spahn1,2, Mike Heilemann1,3

Contact email: [e-mail]

Affiliation(s): 

1) Institute of Physical and Theoretical Chemistry, Max-von-Laue Str. 7, Goethe-University Frankfurt, 60439 Frankfurt, Germany

2) ORCID: 0000-0001-9886-2263 

3) ORCID: 0000-0002-9821-3578