Characterizing glycosylation regulatory networks using single-cell multiomics and mathematical modeling

Glycosylation is a ubiquitous post-translational modification in mammalian systems that fine tune or fully control every aspect of cellular function. This process involves the biosynthesis of glycans by the concerted action of ~400 genes that are called 'glycogenes'. Because of the importance of glycosylation in normal physiology and pathophysiology, understanding how glycans are regulated is of the utmost biomedical importance. The factors that regulate glycan biosynthesis in mammals remain incompletely known because systems-level characterization of glycosylation regulatory network is absent. This project brings together synergistic expertise in systems biology and bioinformatics (Gunawan), glycobiology and biomolecular engineering (Neelamegham), and machine and deep learning (Chen), to address this knowledge gap. We hypothesize that single-cell profiling coupled with mechanistic and deep learning-based modeling and analysis can reveal the key regulators and regulatory networks of glycosylation. The specific aims are: 1) Generate single-cell epigenomics, transcriptomics, and glycomics profiles in hematopoietic stem cell (HSC) differentiation. We choose blood cell system due to their broad biological importance and ease of access. This aim produces single-cell multi-omics data related to glycosylation that will be computationally analyzed in subsequent aims using mathematical and deep learning (DL) modeling. 2) Curate transcriptional regulators and reconstruct gene regulatory networks of glycosylation using data mining and integrative bioinformatics analysis of single-cell data. This aim focuses on transcriptional regulation of glycogenes. We will catalog transcriptional regulators (TRs) and reconstruct and experimentally validate gene regulatory networks of glycosylation in HSCs using single-cell epigenomics and transcriptomics data. 3) Bridge the expression of glycogenes and glycans using flux analysis and deep learning to elucidate regulatory factors of glycosylation. This aim employs first-principle and deep learning models of glycosylation reaction networks to learn the complex, non-linear mapping from glycogene expression to glycoenzyme activity to glycosylation fluxes and glycan abundances. A novel DL model using a combination of Representation Learning and Graph Attention Network will be developed. Systems analysis of the model using Metabolic Control Analysis will provide network-level insights on the regulators of glycosylation. Model-derived glycosylation regulators will be experimentally validated in HSCs, the data from which will be used to fine-tune the model. Overall, this project will generate systems-oriented methods for the Glycosciences that will enable linking cellular epigenetics, transcriptomics, glycoenzyme activity, glycosylation network, and glycan structures. By iterating modeling, systems analysis, and experiments, we will generate insights into gene-level and network-level regulation of glycosylation. Given the importance of glycosylation in human biology, such insights will have broad impact on basic science and disease studies, and in the development of related protein therapeutics.