xCell output and immune cell ratio

Question

[a.stef.44]

Hi everyone,

I calculated immune cell composition using the xCell immunedeconv method, but I realized the scale is quite large, approximately from 1e+18 to 1e-18. I would like to calculate immune cell ratios and compare them between case and control groups. I ran the Mann-Whitney test, but as expected, with these starting values, the ratios were extremely large, leading to downstream statistics and log fold change values in the range of millions.

I calculated lfc as: LFC= Mean(case) - Mean(controls)

Is there a way to normalize the data before calculating ratios and performing statistics to obtain more reasonable and interpretable results?

[yura.grabovska]

What data are you deconvoluting, does the data look ok before running deconvolution. Have you tried running some other test data, does it look the same afterwards?

[a.stef.44]

I done deconvulation on TPM normalized counts data. On the same data I've done TIMER, MCP, CIBERSORT, but non of these shown the same problem. Even xCell had a good results before calculating ratios.

xCell output before statistics

xCell statistical results

But when I calculated ratios between immune cells i got following results, with huge differences in terms of lfc, variations etc. I would like to highlight that I calculated ratios as well for CIBERSORT,MCP and TIME output and I done statistical analysis using the same code.