Dataset title: UT-GPCR002 Machine learning models for CHO-K1 cell segmentation from fluorescence and bright-field microscopy images Authors: Tahk, Maris-Johanna; Torp, Jane; Ali, Mohammed A.S.; Fishman, Dmytro; Parts, Leopold; Grätz, Lukas; Müller, Christoph; Keller, Max; Veiksina, Santa; Laasfeld, Tõnis; Rinken, Ago Contact information: laasfeld@ut.ee; ago.rinken@ut.ee Chair of Bioorganic chemistry, Institute of Chemistry, University of Tartu General dataset description: The dataset was gathered as a part of Tahk MJ, Torp J, Ali MA, Fishman D, Parts L, Grätz L, Müller C, Keller M, Veiksina S, Laasfeld T, Rinken A. Live-cell microscopy or fluorescence anisotropy with budded baculoviruses-which way to go with measuring ligand binding to M4 muscarinic receptors?. bioRxiv. 2021 Jan 1. https://doi.org/10.1101/2021.12.22.473643 The machine learning models in this dataset are provided as zip files with one model file in each zip file. Ilastik models The dataset contains three Ilastik models for CHO-K1 cell segmentation. The models were trained to work on images at resolution of 0.3225 µm/pixel All these models are based on the random forest algorithm implemented in Ilastik software. Ilastik software can be obtained from https://www.ilastik.org/download.html Ilastik version 1.3.3post3 was used for generating the models. In the model names RF stands for "random forest", BF stands for "bright-field" and FL stands for "fluorescence" Ilastik model descriptions RF-FL-1 - a random forest based model for segmenting DiI (a fluorescent membrane label) stained CHO-K1 cells from fluorescent images from the RFP (red fluorescent protein) spectral range. The model expects a uint16 tiff image file as an input with axis as yxc (904, 1224, 1). Other x-y input sizes probably work, but have not been tested. The model produces an hdf5 image file as an output which contains the simple segmentation map as an uint8 value (pixel value 1 corresponds to cell, value 2 corresponds to background). RF-BF-1 - random forest based model for segmenting CHO-K1 cells from contrast enchanced images generated from bright-field Z-stacks. The model expects a uint8 tiff image file as an input with axis as yxc (904, 1224, 1). Other x-y input sizes probably work, but have not been tested. The model produces an hdf5 image file as an output which contains the probability map as a 32 bit float value with three channels Channel 1: Near membrane background (NMBG) pixels Channel 1: Intracellular (IC) pixels Channel 3: Membrane (MB) pixels RF-BF-2 - random forest based model for segmenting CHO-K1 cells from contrast enchanced images generated from bright-field Z-stacks. The model expects a uint8 tiff image file as an input with axis as yxc (904, 1224, 1). Other x-y input sizes probably work, but have not been tested. The model produces an hdf5 image file as an output which contains the probability map as a 32 bit float value with four channels Channel 1: Near membrane background (NMBG) pixels Channel 2: Intracellular (IC) pixels Channel 3: Membrane (MB) pixels Channel 4: Background (BG) pixels Keras models The dataset contains two U-Net3 based Keras models for CHO-K1 cell segmentation. The models were trained to work on images at resolution of 0.3225 µm/pixel. Both models are based on the U-Net3 algorithm implemented in Ilastik software. In the model names RF stands for "random forest", BF stands for "bright-field" and FL stands for "fluorescence". Keras model descriptions U-Net3-FL-1 - U-Net3 based model for segmenting DiI (a fluorescent membrane label) stained CHO-K1 cells from fluorescent images from the RFP (red fluorescent protein) spectral range. The model input should be a 288x288 matrix scaled between 0 and 1. The model output is a 288x288 matrix representing the probability map representing the probability of a pixel being cell instead of background. U-Net3-BF-1 - U-Net3 based model for CHO-K1 cells from a single in-focus bright-field image. The model input should be a 288x288 matrix scaled between 0 and 1. The model output is a 288x288 matrix representing the probability map representing the probability of a pixel being cell instead of background. Methodologial information: All details of model training are given in the accompanying publication available at https://doi.org/10.1101/2021.12.22.473643 Model inference has been validated using Aparecium software available at https://github.com/laasfeld/Aparecium and https://gpcr.ut.ee/aparecium.html Ilastik models can be viewed using Ilastik software Ilastik software which can be obtained from https://www.ilastik.org/download.html Data specific information: Abbreviations: BF - bright-field RF - random forest FL - fluorescence License information: "UT-GPCR002 Machine learning models for CHO-K1 cell segmentation from fluorescence and bright-field microscopy images" is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/. Funding: This dataset was supported by the University of Tartu ASTRA Project PER ASPERA, financed by the European Regional Development Fund, by the Enterprise Estonia Applied research programme 2021, financed by the European Regional Development Fund, by the Estonian Research Council grant (PSG230), by the COST action CA 18133 ERNEST, by the Research Training Group GRK1910 of the Deutsche Forschungsgemeinschaft (DFG), Wellcome (206194), and the Estonian Centre of Excellence in IT (EXCITE) (TK148).