orenbenkiki
commented
3 years ago The full answer for this is that we are working on a vignette that demonstrates the iterative process - taking raw data, generating MCs, figuring out genes that need to be excluded/forbidden, repeating this until we reach "high quality" annotations. However due to vacations etc. this won't be available for a few months.
The short answer is that the "forbidden" (can't be "feature") genes are genes that describe biological processes which are not relevant to the question being investigated - e.g. cell cycle genes are irrelevant for finding "cell types". These genes are never used to create metacells but we still try to ensure that each metacell is reasonably uniform in their expression level (that is, we will try hard to separate, say, T-cells from B-cells, but having done that, we'd still try to ensure each T-metacell or B-metacell has a single cell cycle state).
In contrast, fully "excluding" genes (taking them out completely) means we don't care at all whether the cells in each metacell have the same or varying expression level of these genes. Also, excluding genes impacts the "total UMIs" we use for viewing gene expression as a fraction. We typically exclude all mitochondrial genes and genes which are highly associated with them so that the "total UMIs" we look at are actually "total except for mitochondria".
hurleyLi
cartographerJ
Hi, Thanks for the tutorial and I was able to run through it. I just had a question about the excluded genes. I wonder whether there is any reference for removing the genes in your tutorial
'IGHMBP2', 'IGLL1', 'IGLL5', 'IGLON5', 'NEAT1', 'TMSB10', 'TMSB4X'. Also, what should be the rationale of removing genes from the beginning (e.g., outliers, extremely high expression, genes known to be biased from single-cell experiment?) Thanks! Hurley