Clarification on DE outputs, population inference, and contrasts in run.CSIDE.replicates

[EmanuelSoda]

Hi,
Thanks for the amazing tool.
We are working with a large cohort of Visium spatial transcriptomic samples across three conditions (A = control, B, and C). Our ultimate aim is to test for cell type–specific differences in gene expression across conditions (relative to control A).

To do that, we have seen that the approach is to fit an intercept only model mention here (#87 (comment))

rctd.reps <- run.CSIDE.replicates( rctd.reps, de_mode = "general", X.replicates = X.replicates, explanatory.variable.replicates = NULL, cell_types = c('cell_types_1', 'cell_types_2'), gene_threshold = 5e-5, fdr = 0.01, weight_threshold = 0.1 )

We have a few questions about how to interpret and extend this:

DE results output
In the rctd.reps DE results, each gene/cell type has associated Z-scores and logFC. Could you clarify:
• Are the Z-scores estimates of mean expression?
• How exactly should the logFC values be interpreted in this context?

Population inference and contrasts
We noticed the population parameter in run.CSIDE.replicates() (which calls CSIDE.population.inference()).
• If set to TRUE, should this be run separately per condition (A, B, C)?
• Is there a way to obtain a logFC per gene per cell type across conditions (e.g. B vs A, C vs A) within a single model?
• More generally, is there support for specifying a contrast matrix, so that we can directly test comparisons of interest (like in standard linear modeling frameworks)?

Regional/segmentation analysis
Each donor tissue section can also be subdivided into spatial regions (segmentations).
• What would be the recommended approach for testing differential expression across conditions within regions (per cell type)?
• Should regions be treated as additional grouping variables in run.CSIDE.replicates(), or would you suggest a different setup?

Any guidance on the correct interpretation of the outputs and the recommended workflow for these extensions would be really helpful.

Thanks a lot for your time and for developing this package!
Kind regards,
Emanuel

[VeBeKay1]

Hi Emanuel,

have you been able to run this command that you mention above?

rctd.reps <- run.CSIDE.replicates( rctd.reps, de_mode = "general", X.replicates = X.replicates, explanatory.variable.replicates = NULL, cell_types = c('cell_types_1', 'cell_types_2'), gene_threshold = 5e-5, fdr = 0.01, weight_threshold = 0.1 ) ?

I also have a case/control design and want to have an "intercept only" model. I've also opened an issue (No. 61) on here because I can't get rid of the error message about cell type numbers - I get the same error if I use the vignette data.
Is it possible that the "intercept only" approach only works on individual samples and not with run.CSIDE.replicates()?
Or is there something that I might have missed and you came across when you did this?

myRCTD <- run.CSIDE(myRCTD, X, barcodes, gene_threshold = .01, cell_type_threshold = 3, fdr = 0.25, weight_threshold = 0.8)
does seem to work, with X just containing a vector of "1s", so maybe I'd need to merge my replicates after I do run.CSIDE() on each individual replicate?

Any insights greatly appreciated!
V.