Heterogeneous gene expression within tissues can be difficult to tease apart. Measuring expression in single cells is associated with high technical noise, and a lot of sampling is required to recover the features of the population. As an alternative, Bajikar et al. combine the robustness of gene expression measurements from random samples of ten pooled cells with computational deconvolution. They generate probabilistic models based on known features of transcription and use maximum-likelihood inference to estimate regulatory states at the single-cell level. Using the approach, they show that a cell state found in less than 3% of the population is needed for three-dimensional culture of breast epithelial spheroids. For limited sample sizes, the approach is more accurate than taking single-cell measurements.
About this article
Cite this article
Unmixing gene expression in cell samples. Nat Methods 11, 229 (2014). https://doi.org/10.1038/nmeth.2863