Quantitative Microscopy Pipeline for Building a Model of the Human Cell

Abstract

We aim to replace the typical textbook artistic rendering of a cell with a model created in silico, based on large datasets of 3D images of genome-edited live cells collected by light microscopes. We have created about 20 human induced pluripotent stem cell lines expressing endogenously EGFP tagged proteins that localize to the major cellular organelles. We plate and feed the cells, using a robot, in 96-well glass bottom plates and image them using spinning disk microscopy. So far, we have collected 3D images from more than 15,000 live cells, comprised of high replicates for each genome-edited cell line. We can also apply drugs externally or store the sample overnight in an incubator. We can prescreen the sample on a slide scanner and then image areas of interest with higher resolution and better image quality on one of the spinning disk microscopes. As a part of our workflow, we wrote software to identify the same sample area repeatedly and on different microscopes. This will allow us to revisit the same location during long-term time-lapse experiments after imaging other cells. The software is compatible with any microscope brand and is thus vendor independent. Only a thin communication layer between our software package and the macro interface of the vendor software needs to be updated to adapt the software to other systems. We are using machine learning approaches to combine this information into a predictive model for protein localization. We will present the outline of the pipeline and the components involved. We will also present our workflow quality control criteria, and the methods we have developed to ensure day-to-day consistency between our data sets. All our procedures, tools and data are shared on our webpage www.allencell.org.