• Current Projects
• Completed Projects

Projects

Deep Cell

Deep learning-enabled software for single-cell analysis

PI: David Van Valen (Division of Biology and Biological Engineering)
SASE: Tom Dougherty, Scholar, Kevin Yu, Scholar, and Shenyi Li, Scholar

Single-cell imaging captures the temporal and spatial variations that underlie life, from the scale of single molecules to entire organisms. Recent advances in multiplexed imaging have enabled simultaneous measurements of many proteins in situ, offering unprecedented visualization of cell types and states in tissues. Despite the richness of information offered from such data, tasks such as identifying the type of individual cells remain highly limited by labor and time. The Van Valen Lab's software suite DeepCell deploys a variety of deep learning models trained to perform single-cell analysis tasks at levels equal or superior to humans, from segmenting cells in pathology images to tracking individual cells in live-cell timelapses. These models render difficult analyses routine and empower new, previously impossible experiments across the life sciences field.

However, large scale, high quality labeled datasets are needed to train such models that perform and generalize across tissues and cell cultures. To this end, the Schmidt Academy worked with the Van Valen Lab to maintain and develop its data-labeling web application, DeepCell Label, expanding its feature set to allow for cell type annotations used for human-in-the-loop workflows. All within a React-based web interface, users can now perform a fleet of tasks, including adding cell type labels to individual cells, plotting and gating protein expression levels, and automatically training a neural network in the browser to learn from a small set of labels and rapidly label the remainder of the image. Such features enable annotation of cell types (or any general semantic labels) at a previously impossible rate.

DeepCell Label provides the tooling that drives the Van Valen Lab's data annotation efforts, including labeling data from scratch, correcting model predictions, and quality controlling the contents of training datasets. Rearchitecting and extending DeepCell Label enables higher throughput and quality data annotation and opens the door for new data labeling tasks that power the next generation of deep learning models for image analysis.

GitHub repository
label.deepcell.org