Hello! Looking for feedback, independent validation, and potentially to warn others. TLDR: Analysis of biological sex-specific gene expression indicates male and female cells mixed within each sample id (at least in our hands) When pseudobulking gene expression by sample IDs, female-specific gene expression (XIST) and Y-chromosome genes (EIF1AY) co-occur, and in the raw counts, are directly correlated with each other. This is biologically impossible, and can't be attributed to batch effects. We had two separate people get the same results to cross-validate this concern. The methods section states that demuxlet was used to demultiplex samples within each batch - it's fea[](url)sible that demuxlet failed and sample IDs were scrambled, leading to cells from biologically male and female samples being assigned to the same sample ID. This hypothesis is partially support by the fact each batch appears to have a relatively homogeneous number of cells expression XIST and EIF1AY. In other words, each batch had a different number of male and female subjects, and thus if demultiplexing failed, the resulting "sample IDs" would thus contain a relatively similar ratio of male and female cells, depending on the batch. This explanation appears to explain the data well, but this is just conjecture however. I've attached plots to show what we are seeing. Demuxlet would need to be re-run and evaluated, with the donors' genetic backgrounds, which we don't have access to. So we aren't able to re-analyzing this dataset with confidence at the moment. We've gone through the typical channels to discuss this, and I think they are reviewing it. I just wanted to post this for the general community as a warning (and also to see if others have see this as well). The dataset is still obviously quite useful for training models, celltyping, etc... but I would caution people from deriving biological insights from the data, given that we do not know which cells were derived from which donors. If anyone has alternative explanations, a work around, or is unable to replicate our results, it would be great to hear from them!  Edit: The ability to embed images of graphs doesn't seem to be working on this forum. I've attached a link to Imgur so that people can view the plots. https://imgur.com/uADtkVY https://imgur.com/v2WEUBA https://imgur.com/fN3Fy5U