I have a brief question regarding the new package limpa. First, of all thanks for the great effort of bringing this forward, I was waiting for this since the Neither random nor censored Bionformatics manuscript.

Reading limpa, it is not clear to me if one should still gain for TREND=TRUE in eBayes after the dpcDE or if it will automatically detect the DPC linear model and would use the extra information of the variance?

In DEqMS I would set the TREND=FALSE as afterwards one would add the peptide count and then run spectraCounteBayes to adjust. Or since limma ~3.52, run a DEqMS like analysis directly by using the peptide/features count as TREND = Peptide.Count on eBayes().

The limpa vignette example uses the default, eBayes(fit). On the limpa manuscript it is written that all limma analysis included the trend verison. ...limma-trend option is always adopted whenever limma is used throughout this article... and was wodering if that included as well after dpcDE.

Thanks a lot!

Dear Alejandro,

Thanks for your interest.

No, one does not set trend=TRUE when running eBayes() after dpcDE(). The variance trends are already incapsulated into the precision weights and hence have already been accounted for when the linear model is fitted. You do not need to estimate a variance trend again. The situation is closely analogous to the voom pipeline for RNA-seq.

In the manuscript, when we say "limma-trend option is always adopted whenever limma is used throughout this article", we mean when implementing competing imputation methods. We used limma-trend for competing methods to try to put them as far as possible on the same terms as limpa. The Methods section says: "For all existing pipelines, we use the limma-trend approach where protein-wise variances are squeezed towards a global mean-variance trend rather than assuming a constant prior variance."