scDRS analysis on a large single-cell RNA-seq dataset

[AI-10]

hey,

I am trying to run scDRS on a single-cell data containing three million cells, but the analysis failed due to server memory limitations. Are there any alternative ways to proceed with the analysis?

I plan to perform 50 random samplings of 5% of the cells for analysis, but I’m not sure if this is reasonable.

[martinjzhang]

Sounds like a good plan. scDRS memory requirement is roughly 3X the size of the data. You may consider randomly splitting your into into X parts, run scDRS on each part, and then aggregate the results.

[AI-10]

Thanks for your timely reply

I'm considering how to combine the results of multiple parts? How do I get a combined Z value and P value? Can I average the Z-scores for the different parts?

Best regards,
Gong

[martinjzhang]

You can directly concatenate the results for the individual-cell level analyses.
For the group-level analysis, you can use Fisher's method (for a given cell type across subsets of data) to combine p-values.