(329f) Integrating Multi-Omics Datasets with Robust Penalized Regression Identifies Context-Dependent Signaling Networks

Cells integrate signals from different external stimuli to decide their fate, such as growth or death. The “context” of a cell – the extracellular and intracellular environments – dictates the structure of signaling networks that determine such cellular responses. Identifying context-specific edges is largely understood as important for predicting and understanding cell decision making logic, however, robust identification of context dependent network structures remains a broadly unsolved problem. We hypothesized that context-dependent network structures could be generated through a combination of robust penalized regression modeling with leave-one-group-out (LOGO) analysis applied to a dataset that captures multiple omic-level dimensions. The LOGO analysis encompasses excluding data from one ligand condition out in the regression model construction and enables detecting ligand dependent associations. The unique dataset we utilized to test this hypothesis was recently generated by NIH-LINCS Consortium, and profiled MCF10A cells (non-transformed, epithelial, breast cell line) with proteomic (proteins), transcriptomic (mRNAs), and epigenetic (chromosomal states) data in response to seven different ligand or ligand combination stimulation conditions (including basal control) that differentially modulate cell proliferation. We then employed the robust lasso-based inference model with LOGO to integrate the three data types and uncover ligand specific sub-networks controlling cell fate decisions. The statistical associations between levels of proteins, phospho-proteins, mRNAs, and transcription factor binding sites on chromosomes are treated as edges of the networks. The incorporation of the LOGO analysis in our computational framework enabled us to find the context dependent sub-networks by comparing the ligand LOGO case to the full data analysis. First, we focused on identifying Interferon-γ (IFNγ), an antiviral and cell death promoting cytokine, dependent networks. Our IFNγ-LOGO analysis revealed two mechanisms in PD-L1 (programmed cell death-ligand 1) expression control through spindle protein FAM83D and histone protein HIST2H2AA3. Another interesting finding of our network-inference model is IFNγ dependent regulation of ACE2, the receptor responsible for SARS-CoV-2 cell entry, through IRF1 (interferon regulatory factor 1) expression. Overall, our novel analysis pipeline integrates multiple data types to delineate context specific signaling sub-networks important in cell growth decision making.