COMET with Integrated Data? Normalization Method suggestions?

[cohenp05]

Hello,

Thank you for developing a great package filling an important niche within the world of scRNAseq. I'd like to use COMET on my dataset and I have two questions. 1) Do you have any recommendations for normalization method to use for input to COMET? I am using 10X data normalized with Seurat's SCTransform, but I wasn't sure if SCTransformed data would be compatible with COMET or if instead I should use log normalized or another method. on an integrated dataset (implemented with Seurat) to identify markers for cell types. 2) Do you have any experience or know of any users using COMET on integrated/batch corrected data, and similarly do you know if COMET is compatible with this kind of data? Would you recommend instead running COMET on single samples to circumvent this issue?

Thank you so much for your help!

Best, Phil Cohen

[ktyssowski]

Did you have any more insight? I am also wondering if I can use COMET on integrated data.

[oshahid]

Hi Phil and ktyssowski,

Just to catch everyone up to speed, Phil sent me a separate email some time ago, to which I replied with the following information:

Hi Phil,

Generally, my understanding is that the counts generated from SCTransform replace the traditional NormalizeData, ScaleData, and FindVariableFeatures functions and can be used as a general normalization technique. In my experience, using SCT counts from individual samples as well as SCT counts from integrated samples will both work fine. What we did was we took CD8 T cells from several different human samples, and after clustering them and finding DE genes, we used the same counts and ran them through COMET. The results were pretty compatible with what we expected from DE gene analysis and clustering anyway.

In general, you can use the same counts that you used to run DE gene analysis and it should be okay.