Single-cell cytometry via multiplexed fluorescence prediction by label-free reflectance microscopy

Abstract

Traditional imaging cytometry uses fluorescence markers to identify specific structures, but is limited in throughput by the labeling process. Here we develop a label-free technique that alleviates the physical staining and provides highly multiplexed readout via a deep learning-augmented digital labeling method. We leverage the rich structural information and superior sensitivity in reflectance microscopy and show that digital labeling predicts highly accurate subcellular features after training on immunofluorescence images. We demonstrate up to 3× improvement in the prediction accuracy over the state-of-the-art. Beyond fluorescence prediction, we demonstrate that single-cell level structural phenotypes of cell cycles are correctly reproduced by the digital multiplexed images, including Golgi twins, Golgi haze during mitosis and DNA synthesis. We further show that the multiplexed readout enables accurate multi-parametric single-cell profiling across a large cell population. Our method can dramatically improve the throughput for imaging cytometry towards applications for phenotyping, pathology, and high-content screening.

Image3C, a multimodal image-based and label-independent integrative method for single-cell analysis

“…Automated classification of cells using convolutional neural networks (CNN, machine learning method specialized in image recognition and classification) has become a promising approach for accurate high-throughput cell analysis that is free from observer bias (Blasi et al, 2016;Eulenberg et al, 2017;Kobayashi et al, 2017;Lei et al, 2018;Nassar et al, 2019;Suzuki et al, 2019). To date, CNN-based automated clustering and classification techniques require preexisting knowledge about the organism or cell type of interest (e.g., cell specific morphological traits within an image set) or the availability of cell-specific reagents (e.g., antibodies), or genomic sequence (e.g., single-cell sequencing) (Table 1 shows an overview of the existing methods) (Baron et al, 2019;Blasi et al, 2016;Cheng et al, 2021;Eulenberg et al, 2017;Hennig et al, 2017;Kobayashi et al, 2017;Lei et al, 2018;Nassar et al, 2019). This means that to make effective use of artificial intelligence (AI) approaches for single-cell analysis, one must have information available to train the algorithm or for machine learning (ML) models, which often arises in the form of information gleaned from the use of reagents like antibodies.…”

“…The combination of the clustered cell images, the outputs of their functional assays and the published literature about closely related organisms might allow the identification and description of cell types of interest. In comparison to existing label-free cell clustering methods, Image3C does not require initial antibody staining (Cheng et al, 2021;Hennig et al, 2017;Lippeveld et al, 2020;Nassar et al, 2019), pre-existing knowledge of specific cell morphology (Suzuki et al, 2019;Yakimov et al, 2019) and is not limited to a specific cellular phenotype (Blasi et al, 2016) for a priori identification of certain cell types (Table 1). This makes Image3C extremely versatile and applicable to virtually any research organism and tissue from which…”

Interpretable deep learning uncovers cellular properties in label-free live cell images that are predictive of highly metastatic melanoma

“…Recent machine learning studies have impressively demonstrated that label-free images contain information on the molecular organization within the cell (Cheng et al, 2021;Christiansen et al, 2018;Guo et al, 2019;LaChance and Cohen, 2020;Ounkomol et al, 2018;Sullivan and Lundberg, 2018;Yuan et al, 2019). These studies relied on generative models that transform label-free images with fluorescent images, which can indicate the organization and, in some situations, even the relative densities of molecular structures.…”