What are the expected upper limits of standard error values when using the limpa pipeline?

I am analysing proteomics data from a control cohort and was surprised to see a number of standard error values just under 4. At a glance, the proteins with the highest standard error appear to be reasonably well detected (~7-16 peptides, PropObs around 0.6 - and comparatively, detected in all samples in the DIA-NN protein matrix). The protein values do appear to be quite variable amongst the cohort.

What do you think could be driving these high standard error values?

Hi Julia,

I am going to assume that you mean standard errors (SEs) returned by dpcQuant(), because later stages in the limpa pipeline do not provide standard errors explicitly. The SEs are attached to individual protein quantifications in individual samples, not to the protein in general. It is not meaningful to refer to a standard error for the protein itself.

You mention the DIA-NN protein matrix, but DIA-NN detects precursors rather than proteins and limpa doesn't make any use of the protein matrix from DIA-NN.

In general, a protein-sample quant value will get a large SE if all or almost the peptides for that protein are missing for that sample. The SE can also be large if the peptides for that protein are inconconsistent with one another, which would tend to make all the SEs largish for that protein.

Anyway, the size of the SEs should not cause any difficulties. The SEs are not used directly in an absolute sense. They are instead used to drive the variance trend modelling done by dpcDE(). The standard errors used by dpcDE() for the final differential analysis are reestimated from between-sample variabiltiy of the protein quants.

If you would like more detailed commentary on a particular example, you would need to provide more specific information about that example.