Single-cell RNA sequencing (scRNA-seq) is revolutionizing how scientists study individual cells, especially how genes interact with each other under different conditions. But understanding those interactions isn’t always straightforward, especially when cells come from multiple individuals and experiments use different methods.

To tackle this challenge, researchers at the University of South Carolina have developed a powerful new statistical tool called scCOSMiX. This method improves how we detect changes in gene-gene interactions by accounting for variability between individual subjects and the structure of single-cell experiments.

Droplet scenario: Precision-recall curves for differential co-expression, evaluated on data simulated based on data sets generated using scDesign3 with the droplet-based data set GSE266919 as a template.

Data sets were simulated using scDesign3 based on a subset of 1225 gene pairs from the template data set.  of the gene pairs were randomly selected and manually given a differential co-expression of  while the remaining 1085 gene pairs have no differential co-expression.

What sets scCOSMiX apart is its ability to model not just whether genes are turned on or off, but how strongly they are working together (or co-expressed) under different conditions. The model also handles complex features of scRNA-seq data, such as “zero inflation” (many genes appear to be off) and differences in how cells are processed.

To test the method, the researchers ran a series of simulations and applied scCOSMiX to real-world datasets (GSE266919 and GSE108989). These datasets come from different scRNA-seq technologies, showing that scCOSMiX is flexible and broadly applicable.

By offering a more accurate way to uncover how genes interact in response to disease, drugs, or environmental factors, scCOSMiX could help researchers pinpoint targets for treatment or better understand cellular behavior across individuals.