5

I would be very grateful if somebody could sketch out the methods Pfam and PANTHER use to assign a family to a given protein and how they are different. My (cursory) understanding is that InterpProp pools databases like Pfam and PANTHER and provides a meta-classification based on some combination of their data.

In particular, I wish to understand where the discrepancy comes from described in the picture below?

enter image description here

I'm fairly new to the ecosystem, so feel free to throw any bone this way.

M__
  • 12,263
  • 5
  • 28
  • 47
rtviii
  • 364
  • 1
  • 7
  • 1
    That is a low homology sequence where different thresholds are applied. Interpro is more narrow than PFam. Also, at a glance looks like a 3 residue repeating pattern (helix) featuring tryptophan and leucine, but it has two prolines in it, so it's probably a linker that sticks to the side of the protein... – Matteo Ferla Apr 17 '20 at 05:14
  • 1
    Hey, @MatteoFerla! I hope you've been well. I'm following up directly on your suggestion from once upon time. I ended up using PFAM to convert to my "yet another" nomenclature in the prototype, but as you said, some repeat proteins get assigned multiple families, which muddies the water a bit. On top of that i've been fiding discrepancies between the families pfam's api provides and their static databases and then there also some proteins which are already classified on InterPro(with pfam as source) but have no entries on PFAM – rtviii Apr 17 '20 at 05:50
  • That is all to say that I'm super interested in how to make my nomenclature converter more robust and in how these databases/classifiers work( i know pfam is based on HMMs). Say, how can it be that InterPro is narrower than the sum of the databases which it gathers under its umbrella for example. If you know of an overview or can throw me a few clue-words to follow up on, i'd be supper happy. – rtviii Apr 17 '20 at 05:54
  • 1
    Where a domain actually truly ends from a structural point of view is not trivial. The SGC (Structural Genomics Consortium) in their semi-automated pipeline for protein crystallisation actually makes multiple constructs generally cutting the ends by half a dozen residues before or after —it sounds silly way of doing it, but it works... – Matteo Ferla Apr 18 '20 at 09:59
  • 1
    About the narrowness... It is just for personal observation as I do not understand how InterPro works with its "signatures" and its blending of databases. In this case the Interpro annotation is a blend of the PFam and the Panther and the latter is longer. If you get something that is more distant, you may get a PFam but not a Panther, so the InterPro won't cover it. – Matteo Ferla Apr 18 '20 at 10:05

0 Answers0