Potential database curation issue: nuclear protein MAML2 handled as intercellular interaction partner?

[RobStrasser]

Dear LIANA-team,

thanks for the outstanding tool - it keeps on giving!

In my analysis, I've encountered MAML2 (ligand) - NOTCH4 (receptor) signalling after using li.mt.rank_aggregate.by_sample(..., resource_name='consensus') with version 1.5.1 . This seems like an artifact in the curation process as the canonical interaction occurs between the cleaved intracellular domain of the NOTCH4 receptor and MAML2 within the nucleus, regulating transcription as a consequence (see e.g. https://pmc.ncbi.nlm.nih.gov/articles/PMC6507302/ or https://www.reactome.org/content/detail/R-HSA-9013695).

I saw that @dbdimitrov is continuing to curate the list by removing erroneous interactions (#177), so maybe this also is a candidate for removal?

In the same issue, there is a brief summary how the ligand-receptor list has been curated in the first place, seemingly with stronger focus on the receptor side compared to ligand ("(ii) we kept interactions only where the receiver protein was plasma membrane transmembrane or peripheral according to the 51st consensus percentile of the localisation annotations"). Might there be a way to apply a similar way to filter putative ligands, removing all proteins that are not secreted or transmembrane?

Lastly, if this was simply a user error I am happy for any pointers to fix this on my end. For now, I'll simply manually remove it in the Liana pipeline.

Minimal reproducible example:

import liana as li
lr_pairs = li.resource.select_resource('consensus')
lr_pairs.query('ligand == "MAML2" or receptor == "MAML2"')

output:

ligand | receptor
MAML2 | NOTCH2
MAML2 | NOTCH3
MAML2 | NOTCH1
MAML2 | NOTCH4

Thanks and best,
Rob

[dbdimitrov]

Hi Rob,

Thanks a lot for pointing this out. Indeed, I will try to exclude it from future versions! And people from saezlab are already working on providing an improved LIANA+ resource in the future :)

Daniel