Modern single-cell sequencing technologies have granted biologists unprecedented access to gene regulatory networks (GRNs) driving cell differentiation. As large parts of these GRNs continue to be uncovered, it is of increasing interest to study the regions of these networks that are temporally modulated throughout an organism's lifespan. This thesis's primary contribution is the development of a novel method for extracting GRN information from aging time courses of single-cell sequencing data. In particular, this method infers the trajectories that cells take through gene expression space as they differentiate at all sampled ages. With these trajectories, we can compute the probabilities that cells will differentiate into particular cell fates, and use these probabilities to identify the critical genes in the GRNs that determine cell-fate specification at each sampled age. The method is based on extending existing stationary stochastic models for cell differentiation to account for the gradual effects of aging observed across all members of a particular species. With this model, I show that the (normalized) aging population distribution of developing cells can be viewed as a quasi-static curve in the Wasserstein space of probability measures. This characterization allows for the principled design of a novel trajectory inference method that utilizes optimal transport (OT) to both reconstruct and debias cell trajectories across all ages simultaneously. This method, global StationaryOT (gStatOT), is the first of its kind to exploit the global information contained across an aging time course to infer more robust cell paths. I demonstrate on both real and synthetic data that gStatOT can significantly improve the accuracy of reconstructed cell trajectories when data sparsity would normally introduce substantial noise into the outputs of current TI methods. Moreover, I show that the improved, globally supported, cell-fate probability estimates produced by gStatOT enable more accurate identification of the critical genes involved in cell differentiation GRNs.
Cole Boyle (Thu,) studied this question.