Studying the embryonic development of specific regions of the brain can provide crucial insights into the molecular mechanisms that underlie its formation and function, and provides the basis for neurodegenerative disease research efforts. Single-cell RNA sequencing is a powerful tool for studying the transcriptomic events during the embryonic development of brain regions such as the striatum, which plays a critical role in Huntington's disease. However, obtaining enough cells for sequencing from such regions can be challenging and may require pooling tissue from multiple animals of different sexes. 

Pooling cells of different cells adds variability to the data and can potentially confound the downstream analysis. Thus, individual cells must be labeled according to their sex so that this covariate can be properly accounted for during analysis. Doing so removes variability unrelated to the process of interest, prevents potential spurious associations, and enables researchers to study the effect of sex.  

Here, we introduce the first in silico cell sex deconvolution approach that models the probabilities of the observed counts using an interpretable generative model. The proposed approach uses the counts originating from genes in the chrX inactivation complex and in the chrY per cell, as well as the cell library size. Our approach provides the mixing proportions of male and female cells in the population, as well as posterior probabilities and a maximum a posteriori estimate of the sex for each cell, allowing researchers to consider this critical variable during downstream analysis.