Project summary Despite remarkable advances in single-cell profiling, machine learning and systems biology, our ability to exploit these measurements is limited by the lack of an appropriate framework to model and analyze them. In this application, I propose an organic synthesis of experimental technological development, mathematical modeling, and machine learning algorithm innovations to move beyond conventional descriptive and merely statistical analyses of single cells to mechanistic and predictive modeling of cell fate transition over time and space, and across transcriptomic, epigenetic and proteomic levels. Firstly, in order to unveil the regulatory networks that govern the maintenance of stem cells and progenitors, I will extend the dynamo framework that published recently to predict key regulators that stabilize or destabilize cells states, e.g. the hematopoietic stem cell state, via sensitivity analyses of the reconstructed vector field. In addition, I will build upon the current success of predicting a broad range of hematopoietic cell fate transitions with our least action path approach to extend it to study other biological systems, such as pancreatic endocrinogenesis. To validate these predictions, I will continue my ongoing collaboration with Dr. Vijay Sankran’s lab (co-mentor lab) to first implemented metabolic labeling based scRNA-seq with the 10x chromium system and integrate it with perturb-seq that championed by the Weissman lab (my mentor lab) to test the predicted factors’ efficacy in maintaining the HSC state. Second, I will develop new approaches to seamlessly integrate multi-omics and harmonize short-term RNA velocities with long-term lineage tracing. By doing so, we can enable even more accurate modeling of single cell fate transitions that consider lineage-resolved, epigenetic, proteomic kinetics, offered by cutting-edge single-cell genomic technologies and cutting-edge deep learning methods. Lastly, I will take advantage of my early access of mouse embryogenesis dataset profiled with the powerful Stereo-seq through my close collaboration with BGI research to build 3D in silico spatiotemporally models of mammalian organogenesis. I will also train myself to study other state-of-the-art in-situ sequencing approaches, for example the STAR-map method from my collaborator, Dr. Xiao Wang from Broad. Through the K99 phase of this proposed career development plan, I will develop new computational toolkits and further strengthen my experiment skills, both in human hematopoiesis, Perturb-seq and spatial transcriptomics. When combining these new skills with my rigorous training in systems biology, and single cell genomics, I will be better prepared to transition into an independent investigator in a top-tier research university. Undoubtedly, my research and career development during both K99 phase and my transition to R00 phase will be greatly facilitated thanks to the excellent research environment in Whitehead institute, Broad and Harvard stem cell institute. To sum up, my proposed study will pave the road to launch my future interdisciplinary team that aims at building mechanistic and predictive models of cell fate transitions with a focus in human hematopoiesis.

Project Narrative Through an organic synthesis of mathematical modeling, machine learning, and single-cell genomics, this project aims to bridges the unmet gap between the “big data” from single-cell genomics and quantitative/predictive modeling in order to address fundamental questions in mammalian cell fate transitions, and to provide potential avenues for regenerative medicine. In particular, I propose to 1) dissect the mechanisms of mammalian cell differentiation, reprogramming, and maintenance through differentiable deep learning frameworks; 2) integrate multi-omics and harmonize short-term RNA velocities with long-term lineage tracing; and 3) build predictive in silico 3D spatiotemporal models of organogenesis.