Transformation for Cytometry Data

Dear mixOmics team,

I would like to use mixOmics to classify my samples with CyTOF data. With CyTOF, I have counts of cell clusters. Usually, CyTOF count data are analysed by taking the percentage of a certain cell cluster of all cells from the respective samples (such as percentage of CD4T cells in total CD45+ cells; equivalent to total sum scaling?). Does not this mean that cytometry data is compositional and then the appropriate transformation should be applied prior to downstream analysis?

With this in mind, I plan to apply centered log-ratio transformation prior to PLS-DA analysis. What do you think of this approach? Or is it okay just to use the percentage data?

Thank you for taking the time to respond.

Best regards,

Dear @mdmanurung,
If you are considering the proportions as input data, then yes I would recommend using the CLR transformation afterwards. Usually we use the CLR transform only for microbiome data, but I have this feeling it would be also appropriate here. You can try both and see whether the CLR might give ‘better’ results in terms of graphical outputs.


1 Like