_load_samples_metadata scrambles metadata

[dmalzl]

Hi I am using mofax to interpret some MEFISTO analyses and tried to plot the factors. When inspecting the plots I realised that they don't match my training setup. Upon closer inspection of what the mofax code does I found that _load_samples_metadata seems to rely on the assumption that model.samples and model.groups have the same ordering. However, model.samples is a dictionary which has no intrinsic ordering so in my case it happens that the two don't line up and thus when concatenating the two frames that are produced in this function the metadata does not line up correctly and as a result the plots do not resemble my training setup. I suggest to either just use model.groups as a unifying list of keys and iterate over model.samples instead of doing those separate, resetting the index and then concatenate two misaligned dataframes.