Causal inference of common and personalized single-cell gene regulatory networks - PROJECT SUMMARY Gene regulatory networks (GRNs) underpin cell identity, function, and response to external stimulations in human health and disease. Their systematic reconstruction from data not only promises to understand disease mechanisms at the molecular level in basic science, but also directly assists the translational search for therapeutic targets and disease subtypes. An accurate GRN reconstruction requires two driving factors – data and method. For data, single-cell technology has presented a unique opportunity for data volume and cell type specificity, but also suffers from severe challenges in sparsity and scalability. In addition, every GRN is heavily influenced by the genetic and epigenetic variations that differ greatly between individuals and cell types. This limits the utility of most existing GRNs, which were reconstructed from the data of cell lines or very few donors, in primary cells and the whole population. In terms of GRN reconstruction method, causal inference holds great promise in its capacity to accurately identify causation from reverse causation and confounding, and therefore reproducibly estimate perturbation outcomes for therapeutic development. However, mainstream causal inference methods face major challenges in GRN reconstruction, falling short in modeling causal kinetics, feedback loops, measurement noise, and GRN rewiring that are widespread in complex biological systems and measurements. We will overcome the data challenges by reconstructing GRNs for each primary cell type/state using 1) population-scale and 2) atlas-scale single-cell datasets that include numerous individuals of the whole population. This project will be built on and extend our recent novel causal inference framework using stochastic differential equations and probabilistic modeling to address the method challenges. We plan to demonstrate their utility in the therapeutic control of gene expression levels. We will release our methodological advances as open-source software and our causal GRNs as a public resource. We expect them to provide the community with comprehensive molecular network knowledge and perturbation outcome prediction to promote disease mechanism understanding and therapeutic target discovery. This will become a part of our long-term goal in generating a full map of multi-modal causal molecular circuits from publicly available data in single-cell and spatial omics.
