You should just analyze each species as an independent experiment. Analyze each separately for differential gene expression, and then compare homologous genes based on those independent differential gene expression results.

You could use the homology tools at the www.jax.org to map human to mouse, or use tools in R/Bioconductor for that, or there is a somewhat outdated application called GEAX in sourceforge to map homologues (has been out of development for a couple of years, but still works). In R/BioConductor there used to be a tool called probemapper but I do not see it there now? Not sure what happened to it, but you can do it with BioMart, or with the NCBI homologene tool (works with Batch Entrez)

GEAX was developed in the past by former colleagues where I work (the Hamner Institutes) but has been effectively deprecated now (no funds to maintain it). As far as I know It does still work, but may not support your specific array sets if you used the most current Agilent arrays available (and please do not ask me to update it, as that is not going to happen - it is not my tool and I am not responsible for it).

That is how I have handled cross species comparisons - each species is an independent experiment with its own controls. I analyzed each that way, and then only compared after the differential gene expression analyses. You can see what I did in a 2012 paper "Cross-species Comparisons of Transcriptomic Alterations in Human and Rat Primary Hepatocytes Exposed to 2,3,7,8-Tetrachlorodibenzo-p-dioxin" which you can download from my profile here.

Thanks both for your answer, but I think my question was not clear. We already did the statistical analysis in both species. The question is we want to know the gene expression variability in each species, something like a coefficient of variation, and afterward compare the variability. Our results are more robust in mice and we want to know if it's because gene expression is less variable in mouse compared to human (probably). We can't use the coefficient of variation (CV) because we have two different species, two different platforms and negative values in human. So I'm looking for another test of technique to do this. Thanks again.

You could take the absolute values of the log2 normalized expression from each experiment and compute a CV from that. Look just at the control samples, since your estimated differential expression is primarily dependent on the variation in those.

Plot the CV of the absolute value of expression across all interrogation probes in each case, and maybe plot this as box and whisker plots across quintiles (or whichever number you want to split things by, quartiles, quintiles or whatever). Look at the pattern of change in CV across the divisions you've plotted to see if there are any obvious changes in variance across each data set.

You could also transform the normalized expression values of each data set first to a standardize them before looking at variance of the standardized values.

Finally we found a simple solution. We transformed the normalized expression value from both species on a scale from 0 to 100. After that, we calculated for both species the CV with the transformed data.

You might want to look at Sweet-Cordero's work:

"We continue to exploit such cross species comparisons as a tool for understanding cancer pathways and networks."

https://med.stanford.edu/profiles/alejandro-sweet-cordero?tab=research-and-scholarship