Developing a Novel Causal Discovery Framework to Unveil Individualized Cell-Cell Communication Networks - Project Summary/Abstract The overall goal of this study is to develop a computational framework integrating deep learning and causal discovery algorithms to unveil the individualized (or instance-specific) cell-cell communication networks (CCCNs) underlying tissue environment heterogeneity of individual tissue samples. The versatility of this framework allows for studying CCCN among cells in both normal physiological conditions and various disease environments. Differences in cell composition and functional states within individual tissue environments contribute to tissue environment heterogeneity. Due to its complexity, it is a challenging problem to understand how cells in a tissue communicate and influence each other’s cellular state. To overcome such limitation, we hypothesize that a CCCN is a causal network in which changes in a cell's cellular state can causally influence neighboring cells' states through ligand-receptor (LR) signal transduction. We propose a suite of causal discovery algorithms to model CCCNs as causal networks. In particular, we will develop and apply a novel individualized causal Bayesian network (iCBN) framework, which discovers a CBN that best explains the potential causal mechanism of cellular interactions in each instance, e.g., a tissue sample from a developing embryo, a biopsy from a lung inflammation patient, and a cancerous tumor. This enables the discovery of diverse CCCNs exploited in different instances of a tissue. By summarizing the causal interactions present in instances of a subpopulation, we could learn the subpopulation-specific patterns. This information can be utilized to discern variations in the dynamic communication network among different groups (e.g., normal vs. disease states, young vs. old (aging), pre vs. on treatment, responders vs. non-responders). This will enable the discovery of novel drug biomarkers/treatments tailored for a specific subpopulation with a similar pattern of molecular profile. Under the assumption that a cell signaling pathway likely regulates the expression of a module of genes, referred to as gene expression modules (GEMs), our investigation into cell-cell communication (CCC) involves 1) Developing the nested hierarchical Dirichlet process (nHDP) model to analyze single-cell transcriptome data and identify GEMs reflecting transcriptomic processes. 2) Identifying LR pairs transmitting specific signals to cells expressing GEMs. 3) Creating a novel iCBN approach for learning the CCCN within each instance in bulk RNA-seq data or within a spatial domain in spatial transcriptomic data. In summary, our study will enable the investigation of diverse CCCNs exploited by individual instances. We anticipate that the algorithms and software tools we develop will provide a novel causal discovery perspective for investigating individualized CCCNs, identifying the mechanism of heterogeneity, discovering novel molecular targets for drug discovery, and ultimately guiding personalized therapy based on individualized dynamic CCCNs (precision medicine).
