IST Lunch Bunch

Abstract: Deep learning is transforming the ability of life scientists to extract information from images. While these techniques have superior accuracy in comparison to conventional approaches and enable previously impossible analyses, their unique hardware and software requirements have prevented widespread adoption by life scientists. Specifically, they require significant amounts of training data and computational power. In this talk, I relay our group's experience  - both the successes and challenges - in adapting these methods for single-cell analysis. We show that by leveraging crowdsourcing, we have been able to construct datasets with over 250,000 unique cellular annotations. We show that these data enable even simple deep learning models to perform accurate segmentation and tracking of single-cells in live-cell imaging experiments. We also show how these methods enable the accurate quantification of single-cells in spatial genomics experiments, and discuss how they are enabling the construction of an immune cell atlas of solid tumors. Lastly, I discuss our lab's new software, DeepCell (http://www.deepcell.org), for training and deploying deep learning models in the cloud. We show that by scaling compute power to meet analysis demand, we can significantly reduce the time necessary for large-scale cellular image analysis.