Phenotypic Characterization Using Open-Source Deep Learning Tools

Phenotypic characterization allows the study of cellular effects caused by substances or genetic modifications. Here, we demonstrate how deep learning tools can help advance phenotypic characterization using high-throughput microscopy imaging. We focus on a particular example, segmenting and classifying different cellular phenotypes, to demonstrate how an automated image analysis can be performed using open-source software that can be used without advanced deep learning knowledge. In particular, we give in-depth descriptions of how to train and deploy a deep learning pipeline for phenotypic characterization, including troubleshooting when things go wrong in different scenarios.