what are your opinions on using limma::arrayWeights() for a LC-MS/MS proteomics data-set on biological replicates ? I performed stress treatment on six independent biological replicates of Arabidopsis leaves but the treatment was definitely not 100% homogeneous due to fluctuating in stress intensities, so i want to weight the biological replicates, but i don't want to completely exclude single replicates. Samples are isobaric labeled Highlight / Low light (in a SILAC-manner) and transformed and median normalized to log2 values.
I tried lmfit(method = "robust") which gave me much more significant hits than the standard least square fit. But judging on the raw values of the six replicates it provides to much false positives and i don't trust the given p-values.Gordon Smyth mentioned before in some posts he also doesn't really like the lmfit robust method at all and his lab instead prefers robustness (if necessary) on eBayes level.
Well, using eBayes (robust = T and/or trend = T) has very little effect on the p-value significant observations of my dataset.
But arrayWeights appears to be a quite good compromise for my data-set, but I'd appreciate some opinions how you would judge it used on a proteomics study since i haven't found a single publication that is actually using it for this case.
Additional question for experts on arrayWeights():: I'm tuning the "prior.n" parameter for arrayWeights(). Generic is "10" , but i got best results with "40" (as far as i understand this squeezes the array weights more strongly towards equality). However, documentation on this is scarce. Do you think its "legit" to tune on prior.n? or should i confirm any assumptions first?
Thanks for any help Steve