UC Irvine

Cell state transitions can be modeled as critical transitions marked by an increase in gene-gene correlation among a subset of expressed genes. These genes, the cell state transition drivers, are of interest for their role in guiding the cell into a new state in response to some signal. Single cell RNA sequencing (scRNAseq) should allow us to measure these changes in correlation across a transition trajectory. However, the distribution of scRNAseq counts does not fit the expectations that traditional correlation analyses rely on, leading to millions of false discoveries. Here, I develop a method to accurately calculate and statistically validate gene-gene correlations from scRNAseq data. Then, I demonstrate that those same principles can be used to derive statistics for feature selection tasks and that using those statistics leads to better clustering results. Because critical transition theory predicts that correlations among driver genes rise as the cell approaches the tipping point and decay once they enter a new state, transient peaks in correlation should mark candidate transition drivers. Therefore, I propose a method for analyzing gene-gene correlation trajectories which uncovers cell state transition driver genes. This method captures known cell state transition drivers in epithelial to mesenchymal transition and neuromesodermal progenitor differentiation. Finally, I use this method to identify SHH-expression driver genes PITX2 and NPAS3 in the developing chick frontonasal prominence. The importance of these genes is then experimentally validated through knockout experiments. These developments allow for principled correlation analyses of scRNAseq datasets and demonstrate the utility of applied critical transition theory in identifying critical transition driver genes.