We recently performed targeted HPLC-MS to identify bile acids in faecal samples from infected mice and would like to create bar graphs of the peak areas to compare them between two groups (n = 3 and n = 4 in each group). To account for missing values and maximise statistical power while avoiding false discoveries, is it appropriate to use limpa for this type of data? Would a parametric test (t-test) or non-parametric alternative (Mann-Whitney U test) be more suitable here?
Wilcox/Mann-Whitney-U cannot work because it is bound to sample size even when all of group X are lower than all of group Y. At 4 vs 3 the smallest possible p-value is > 0.05, it can never be below typical cutoffs after FDR.
I am not familiar with targeted HPLC-MS and I don't know what you mean by "peak areas". If it produces standard format mass spec data from a standard quantification tool such as DIA-NN or Spectronaut, then limpa would be appropriate, and much better than the alternatives you mention. t-tests or Mann-Whitney would both be very poor with small sample sizes and high rates of missing values. Mann-Whitney doesn't seem an alternative at all, because you can't get results from non-parametric tests for such small samples.
If the number of targeted features is small, so that the dataset has only a modest number of rows, then I would use the default DPC assumed by dpcQuant() instead of trying to re-estimate the DPC from the data. In other words, I would skip the dpc() step in the limpa pipeline.
Wilcox/Mann-Whitney-U cannot work because it is bound to sample size even when all of group X are lower than all of group Y. At 4 vs 3 the smallest possible p-value is > 0.05, it can never be below typical cutoffs after FDR.
p = 0.057