Dear ROBIN, I have been comparing the log2 values and statistical values generated by ROBIN vs by limma (conducted in R). I entered the same settings into ROBIN and limma and nicely got identical log2 values from my test samples (two color microarrays; triplicates of condition 1 vs Ref and of condition 2 vs Ref). However when I generate the moderated t-statistic with ROBIN vs limma, I get different values. (Also different values for P and adjusted P....). It looks like the settings are the same in the ROBIN R script and the limma script, so it doesn't make sense that the statistical values come out different. I am using the feature Mean, background Median, Printtiploess, NormExp, Aquatile, Separate, BH, 0.05 P settings. It doesn't seem to matter that ROBIN is using an older version of limma than our R script. I've also generated sample files with Unique Names so that ROBIN doesn't collapse the values from redundant and control spots, which is something our script does not do. We are absolutely stumped here. We are clearly missing something in the script. Is ROBIN using a default setting for the data it is using to generate the t value? We'd like to use ROBIN for our analyses. We'd like to fully understand the calculations and methodology.

Dear Cathy,

what you describe is surprising, however the only explanation that comes to my mind would
be that either your script is running on a different limma/R version combination and/or your
script and the one generated by Robin are using different implicit default values for limma
functions.

Across limma releases, some function signatures and default values (that are used
if you do not pass any custom values) change and one and the same script can hence
produce a different output. Have you tried running the Robin script both under the R/limma
version embedded in Robin and your installed R/limma? Did you get the same or different
statistical results?

cheers,
Marc