Putative issue with estimating kinase activity based on information from Omnipath db using cancer phosphoproteomics measurements

[saezlab]

Putative issue with estimating kinase activity based on information from Omnipath db using cancer phosphoproteomics measurements

[deeenes]

Dear @Jasonmbg,

Thanks for reporting these likely erroneous database records.
As all databases, the resources constituting OmniPath contain wrong records, some of them more, others less. Here we see all these records are from only one resource called ProtMapper. ProtMapper uses literature mining algorithms (REACH, Sparser), which are known to produce a number of false positives in their output. Seeing the high number of false positives, I might consider removing ProtMapper from OmniPath in the future, or separating from other resources, actually it's authors also recommended this. For you I recommend removing the only ProtMapper interactions from your data frame:

omnipath_ptm %>% dplyr::filter(!grepl('ProtMapper', sources) | n_resources > 1)

Best,

Denes

[deeenes]

Yes. And if you want to further reduce false positives, you can remove all the rest which have no literature references (and not from PhosphoSite -- we might miss some of the PS references). ProtMapper is tricky, bc it has references, but it's false positive rate is high.

[Jasonmbg]

Dear @deeenes,

good morning and thank you very much for confirming this discrepancy; thus, for directly estimating kinase activity analysis, the above code chunk would suffice for removing these false positives, correct?

Best regards,

Efstathios