search

NCATSTranslator / Feedback

A repo for tracking gaps in Translator data and finding ways to fill them.
7 stars 0 forks source link

MVP2 'what may increase activity of human DHODH'... we really need better EPC #916

Open gglusman opened 2 months ago

gglusman commented 2 months ago

MVP2, what drugs may increase expression of human DHODH:

The top answer (score=5) is 5'-(Cyclopropylmethyl)spiro[cyclohexane-1,1'-furo[3,4-c]pyridine]-3',4'-dione. This is a lookup edge: image, supported by one paper and one source, BindingDB.

I looked into the full text of the paper and couldn't really find mention of the molecule. Various related and similar things are mentioned, but I couldn't clearly match it to this molecule.

The EPC link to BindingDB sends me to the wiki page, from which I went to BindingDB's homepage, where I searched for the molecule by name... unsuccessfully. So I used its ChEMBL identifier instead (CHEMBL4795021) - it was offered as a synonym by the PubChem page, as linked from the result in the UI. The BindingDB page indeed claims that this compound targets human DHODH: image The evidence it provides for this assertion is "Curated by ChEMBL", with a link to ChEMBL's homepage. So I went to ChEMBL and searched for this compound (easy given I have the ChEMBL identifier). In the compound report card, I scroll down to the "Target Predictions" section, where I find a small table with 339 predicted targets. Among them, I see: image Since the content of the table is rather obscure (three levels of % confidence, and an unclear Activity Threshold), I went to the suggested paper to educate myself. In the Data Preparation section of the paper, I read that, apparently, there is a "default activity threshold of 6.5 logarithmic value units" for classifying things into active or inactive, which this prediction doesn't meet... so I probably shouldn't give much credence to things labeled 'Inactive'.

So, in summary, after much digging through obscure EPC rabbit holes, the top result seems to be based on a weak prediction, and on results from just one paper - that I'm still not sure even talks about this molecule.

Addendum: There is no indication on which actor provided this result. All I can tell (via the UI) is that it goes away when I filter out results from either Aragorn or Unsecret, but not when filtering out results from imProving or BTE. (But see below...)

gglusman commented 2 months ago

Digging via ARAX interface, I see this from Unsecret: image The two support edges shown appear to be identical.

From Aragorn: image image

So it looks like the two ARA assertions are based on exactly the same source information.

gglusman commented 2 months ago

Just to make more explicit a point that may have been a bit obscure above. BindingDB is tagged in the EPC as primary_knowledge_source, but it in fact points at ChEMBL for its source. It is not entirely clear to me whether ChEMBL is the primary source, either...

sstemann commented 2 months ago

tagging Unsecret and EPC for their review of the findings, which are thorough and much appreciated.