A Single-Cell Interrogation System from Scratch: Microfluidics and Deep Learning.

Abstract

Live-cell imaging using fluorescence microscopy enables researchers to study cellular processes in unprecedented detail. These techniques are becoming increasingly popular among microbiologists. The emergence of microfluidics and deep learning has significantly increased the amount of quantitative data that can be extracted from such experiments. However, these techniques require highly specialized expertise and equipment, making them inaccessible to many biologists. Here we present a guide for microbiologists, with a basic understanding of microfluidics, to construct a custom-made live-cell interrogation system that is capable of recording and analyzing thousands of bacterial cell-cycles per experiment. The requirements for different microbiological applications are varied, and experiments often demand a high level of versatility and custom-designed capabilities. This work is intended as a guide for the design and engineering of microfluidic master molds and how to build polydimethylsiloxane chips. Furthermore, we show how state-of-the-art deep-learning techniques can be used to design image processing algorithms that allow for the rapid extraction of highly quantitative information from large populations of individual bacterial cells.