Dependency on Background Cells (followup)

[NoahMottelson]

Hey again,

In the 7th discussion point of the manuscript you mention the possibility of choosing the control gene sets based on a different data set than the one you're analyzing. Is this implemented as an option in the software yet?

[martinjzhang]

Hi,

There is no implementation yet. But we are working on it. There will likely be an implementation in a month or two.

Best,
Martin

[NoahMottelson]

Thanks for the super fast reply.

[martinjzhang]

Hi Noah,

I am following up on this issue.

We currently think it is a complicated issue. The main problem is that the target scRNA-seq data and the reference scRNA-seq data may have different modalities (e.g., collected using different technologies) and are not comparable. In this case, it is hard to make statements about false positive control of the method.

We did add a new option to adjust for cell group proportions. Specifically, it takes a set of cell group annotations and inversely weight cells by the cell group proportions (done implicitly within the computations of scDRS). This option may partially address the issue of the imbalanced data set. See the "adj_prop" option in https://martinjzhang.github.io/scDRS/reference_cli.html

It may take a while for us to finally add the option for using reference data sets. If it is urgent to your research project, a hacky way is to, after calling scdrs.preprocess, substitute the mean_var column in adata.uns["SCDRS_PARAM"]["GENE_STATS"] with your own 20*20 mean-variance gene bins computed from the reference data (categorical). Then, you should be able to call scdrs.score_cell, which finds control genes based on your definition of mean-variance gene bins.

Let me know if you have further questions.

Best,
Martin