Data Integration, Dynamical Modeling and Statistical Learning to Unravel the Pluripotency Regulatory Network in Embryonic Stem Cells.

摘要

Embryonic stem cells (ESCs) are pluripotent cells characterized by their capability to self-renew and differentiate into any adult cell types. Recent efforts in systematically profiling ESCs have yielded a wealth of high-throughput data. Complementarily, emerging databases and computational tools facilitate ESC studies and further pave the way toward the in silico reconstruction of regulatory networks encompassing multiple molecular layers. In Chapter 1, I review the state-of-the-art databases, algorithms and software tools, with a focus on those applied to ESC studies. These resources are used to organize and analyze high-throughput experimental data collected to study mammalian cellular systems. In Chapter 2, I describe how I constructed a comprehensive, ESC-specific database called Embryonic Stem Cell Atlas from Pluripotency Evidence (ESCAPE) by integrating data from many high-throughput ESC studies. A 30-node signed and directed subnetwork for self-renewal and pluripotency of mouse (m)ESCs was then extracted from ESCAPE. The underlying regulatory logic among subnetwork components was then learned using single cell gene expression measurements together with the initial network topology. Comparison of the learned logic for subnetworks in serum vs. 2i revealed differential regulatory roles for Nanog, Oct4, Sox2, Esrrb and Tcf3. Validated by experiments, the dynamical modeling of the learned subnetwork upon single and combinatorial gene knockdowns revealed that Oct4 has the most significant effect on the pluripotency machinery. In Chapter 3, I applied a pipeline called Expression2Kinases to infer upstream transcription factors (TFs) and protein kinases (PKs) and their pseudo-activity patterns from genome-wide gene expression profiles to globally map the regulatory landscape of mESCs and their differentiated progeny. This approach provided an integrated view of the interrelationship among a growing number of signaling and transcriptional regulators in controlling ESCs fate decisions. In Chapter 4, Support Vector Machine (SVM)-based classifiers were developed to predict genes important for self-renewal and pluripotency of mESCs. The SVM-based predictions benefit from using heterogeneous data types and the RBF-based kernels for training. As summarized in Chapter 5, altogether, the ESCAPE database with the validated dynamical model of the pluripotency subnetwork in mESCs, the pipeline for characterizing the global cell fate landscape and the pluripotency gene classifier improves our current understanding of the molecular transcriptional and cell signaling machinery controlling the pluripotency and early differentiation of ESCs.