Realtime morphological characterization and sorting of unlabeled viable cells using deep learning

Abstract

Although cell morphology is often the gold standard for diagnosis and prognosis of many diseases, and conditions, it has seen limited application in combination with comprehensive molecular and functional characterization. This is largely due to the manual and subjective process of collecting cell morphology information and limited methods for sorting that do not perturb the cells. Here, we introduce the COSMOS platform, which is capable of high-throughput cell imaging and sorting, based on deep learning interpretation of high-content morphology information in realtime, yielding populations of cells that are label-free, viable, and minimally perturbed. We demonstrate the utility of the platform by enriching tumor cells and performing gene expression analysis on sorted cells showing concordance between morphological and molecular assessments.

Artificial intelligence-driven morphology-based enrichment of malignant cells from body fluid

“…The copyright holder for this preprint this version posted January 25, 2023. ; https://doi.org/10.1101/2023.01.24.525423 doi: bioRxiv preprint evaluated cell viability of breast and ovarian cancer pleural fluid samples by trypan blue exclusion with and without instrument processing. Paired t-testing demonstrated no significant differences in viability after instrument processing, which is consistent with our previous results 22 (Fig. 3A).…”

“…1B) . The model training set of images were sent to the deep convolutional neural network for multi-dimensional (>1,000 dimensions) quantitative morphological image analysis and model training, as described previously 22 . A model validation set of images, separate from images used for model training set, was created using 2,031,024 images of both cell lines (MeT-5A) and FACS purified patient-derived cell types, as described above and in Methods ( Fig.…”

“…Many enrichment methods rely on fixation, epitope tags, or other techniques that may alter and destroy cells during the process. We previously reported cell lines and dissociated tumor tissue run through the Deepcell platform retain viability and gene expression profiles 22 , but have not tested clinical effusion samples. We evaluated cell viability of breast and ovarian cancer pleural fluid samples by trypan blue exclusion with and without instrument processing.…”

“…We previously reported cell lines and dissociated tumor tissue run through the Deepcell platform retain viability and gene expression profiles 22 , but have not tested clinical effusion samples. We .…”

Pooled genetic screens with image‐based profiling

“…In addition to microscopy‐based approaches, specialized cell sorters have been developed to reconstruct fluorescence or Raman microscopy images of cells and sort cells on image‐based criteria in real time (Nitta et al , 2018, 2020; preprint: Salek et al , 2022; Schraivogel et al , 2022). These approaches have been demonstrated with diverse phenotypic measurements, including fluorescent reporter localization and surface epitope immunofluorescence (IF) in live cells and intracellular IF in fixed cells.…”