xCell enrichment score normalization #76
Description
Hi,
I have an enrichment score matrix with cell values heavily skewed towards the right. 40-50% of samples have enrichment score cell values which is zero or close to zero (<1e-18). Normalisation of this with conventional methods such as log10 normalisation and using square root of the values does not make the distribution Gaussian. If I remove the samples with enrichment score cell values which are zero or close to zero and then log10 normalise, the distribution does become more gaussian.
However, I don't know if this is the preferred method to do the analysis.
I aim to show significant differences between enrichment scores of different groups of patients and would like to correlate bulk RNA seq enrichment score counts with multiplex immunofluorescence to see if results are somewhat similar. Would it be advisable to remove samples and perform parametric testing after normalisation or keep all samples and perform non-parametric testing?
