Congrats to Sisi Chen, Paul Rivaud, Jeff Park, Tiffany Tsou, and Matt Thomson for their new paper titled: Dissecting heterogeneous cell populations across drug and disease conditions with PopAlign, just published in PNAS.
This paper describes a computational framework in which we use probabilistic models to represent population data from single-cell samples, and use the models to compare samples at multiple levels: gene, cell-state, and whole population. The probabilistic models unlock a host of statistical metrics that allow us to statistically ‘align’ subpopulations across samples, and tag them for downstream comparison. What’s cool about the method is that the models compress the data by 50-100x, so we can compute across hundreds of samples without running into limitations in memory or compute resources.
The code can be found at https://github.com/thomsonlab/popalign, and new tutorials are in the works.