Projecting to cell level

[seanken]

Really great tool (already getting some nice results with it!)! One question: is there a built in way to project the recovered gene programs to the single cell level (so each cell gets a score for each factor instead of each sample)? Know it is not quite what the method is built for but could be useful (both for understanding the factors and for certain downstream analysis). I'm sure I could work something out on my end using the gene loading information, but was wondering if there was an obvious/built in way to do so already (I'm in a setting where the sample level statistic might make less sense than the cell level information)? Any insight would be great, thanks!

[j-mitchel]

Thanks for using scITD, and glad it's working well for you! We haven't implemented this yet, but it's a good idea. It should be fairly straightforward to code, but I would be careful about how you scale expression. Since we typically initially scale the pseudobulk expression (for a given gene-cell type pair) across donors, I would think that it's necessary to use the same mean/sd scaling for transforming (normalized) expression of the individual cells as well.

[seanken]

Thanks for the reply! Good point on the normalization, will definitely take it into account!