High-throughput label-free detection of DNA-to-RNA transcription inhibition using brightfield microscopy and deep neural networks

Abstract

Drug discovery is in constant evolution and major advances have led to the development of in vitro high-throughput technologies, facilitating the rapid assessment of cellular phenotypes. One such phenotype is immunogenic cell death, which occurs partly as a consequence of inhibited RNA synthesis. Automated cell-imaging offers the possibility of combining high-throughput with high-content data acquisition through the simultaneous computation of a multitude of cellular features. Usually, such features are extracted from fluorescence images, hence requiring labeling of the cells using dyes with possible cytotoxic and phototoxic side effects. Recently, deep learning approaches have allowed the analysis of images obtained by brightfield microscopy, a technique that was for long underexploited, with the great advantage of avoiding any major interference with cellular physiology or stimulatory compounds. Here, we describe a label-free image-based high-throughput workflow that accurately detects the inhibition of DNA-to-RNA transcription. This is achieved by combining two successive deep convolutional neural networks, allowing (1) to automatically detect cellular nuclei (thus enabling monitoring of cell death) and (2) to classify the extracted nuclear images in a binary fashion. This analytical pipeline is R-based and can be easily applied to any microscopic platform.

Keywords: Cellular imaging; Convolutional neural network; Immunogenic cell death; Semantic segmentation; Systems biology; Transmitted light microscopy; UNET.