Metadata-aware MBR #2564
Description
Firstly, thank you for all your work of the Fragpipe ecosystem. It's been an enormously valuable resource to me and my colleagues. I know this has come up before, but there is an issue I see myself confronted with regularly:
Currently, MBR is based purely on overall correlation between samples. However, in our data (proximity labelling in plants), we regularly have a scenario where correlations of samples from different baits are stronger than between samples of the same bait. This happens e.g. when we expose our plants to various stressors, which drastically affect the overall background proteome, but might change relatively little in terms of key interactors. That means that even restricting MBR top-runs does not efficiently prevent MBR between groups. MBR then often leads to false positives, where a bait protein is found in a sample of another group, where it is in reality absent.
Perhaps implementing metadata-aware MBR could fix this, e.g. through an additional MBR group parameter, to prevent matching between the different groups? Is there any desire from your end to implement something like that?
Best,
Leonard