UCLA

Recent developments in cell labeling techniques allow studying activities of the massive cell population at single-cell or clone-level resolution. The generated data are usually featured by a large number of labels but small sizes of samples. However, the underlying clonal dynamics are usually stochastic and high-dimensional in nature. Thus, inferring the full upstream mechanisms from the downstream sample data is essentially an overfitting problem and poses new challenges in associated computational, statistical, and modeling methods. In this work, I studied the clonal dynamics in the hematopoietic system of rhesus macaques based on a decade-long clonal-tracking experiment. I first develop a computational algorithm that tries to improve the quality of the sampled data by correcting DNA-sequencing errors. Then, I take the cell count (clone size) data and build a multi-compartment neutral model to study the dynamics of each labeled clone. To avoid overfitting, I simplify the mechanistic model and select robust statistical features of the data. Finally, when analyzing the birth-death-immigration (BDI) clonal dynamics under global carrying capacity, I find that the usually invoked mean-field approach fails to predict simulated distribution of clone sizes. I solve this problem by transforming the problem and further approximating the carrying-capacity effect by the fixed-total-size constraint in a Moran model. I hope this work not only solves the current technical problems, but also contributes to the ongoing efforts in understanding the long-term multi-clonal dynamics in complex systems.