Using the Past to Predict the Future: Applications of High Throughput Longitudinal Single Cell Analysis, Genomics, Stem Cells and Machine Learning to Discovery
MetadataShow full item record
Clinical trial failure rates for neurological diseases are some of the highest of any indication (~93%), despite showing promise in preclinical disease models. The poor predictive power of preclinical models is a major obstacle to therapeutics development. To improve matters, we developed an array of human brain cell models differentiated from patient-derived induced pluripotent stem cells (iPSCs), which exhibit disease-associated phenotypes. To fully harness these models to understand mechanisms of disease and screen for therapeutics, we developed systems to perform high throughput longitudinal single cell imaging and analysis, called robotic microscopes. Automated programs find cells within the images and track them longitudinally increasing the sensitivity of screens 2-to-3 orders of magnitude compared with conventional approaches. To expand the power of the platform, we are developing an array of over 270 biosensors that report a variety of biological structures and functions from cells and circuits. Increasingly, deep learning approaches are used to extract features from images, enabling us to achieve super human performance in some cases. We also are using family-based whole genome analysis of patients with neurodegenerative diseases to find new targets and validate then in patient derived human brain cell models. Observational studies in a model of one neurodegenerative disease made it possible to unravel complex cause-and-effect relationships and to identify the autophagy protein clearance pathway as a potential therapeutic target. We then launched a small molecule discovery and development effort that led to brain penetrant orally available autophagy inducers that mitigate disease phenotypes in our human neuron models of neurodegenerative disease. Combining robotic microscopy, biosensors and deep learning, it is becoming possible to develop dynamic predictive models of cell fate at a single cell level that serve as blueprints for therapeutic strategies.