Comparison of the AA position of domains on cBioPortal and UniProt

Question

[larissa.kluxen]

I would like to use the databases cBioPortal and UniProt to create a list of relevant domains in certain genes. I know that cBioPortal is hg19 annotated and UniProt is hg38, but that shouldn't matter for the amino acid sequence, should it? For example, for ZRSR2, UniProt gives position 198-304 for the RRM domain, while cBioPortal gives 246-302 (and this with the same RefSeq and ensemble annotation). How do I know which information is more credible? And where does the difference come from? Thank you :-) Larissa

Answer 1

[larissa.kluxen]

If someone is interested in the solution, I got a mail from UniProt support:

The discrepancy you refer to is not related to the reference genome assemblies used by UniProtKB and cBioPortal, as you have noted that the sequences in both resources are identical. However, for some genes, the prediction of the encoded protein could differ if different genome assemblies are used. In such cases, discrepancies could arise from using different protein sequences to predict domains. The UniProtKB sequence would then be more reliable as it uses an updated version of the reference genome and the sequence has been curated by an expert. In your example, given the same amino acid sequence, both resources predict an RRM domain but with different ranges. This difference is due to the use of different domain predictors. As you can see from InterPro (https://www.ebi.ac.uk/interpro/protein/reviewed/Q15696/), different predictors give different ranges. ProSite (PS50102) predicts a range between 198 and 304. This is the RRM domain prediction manually selected in UniProtKB/Swiss-Prot as it best fits the functional structural domain. PFAM (PF00076) predicts a range between 241 and 297, while cBioPortal uses another PFAM signature (https://www.ebi.ac.uk/interpro/entry/pfam/PF13893/) that predicts the RRM domain with a range of 246-302. Unfortunately, there is no way to convert one annotation into the other. Depending on your final aim, you might want to use both predictions to get maximal coverage, as one predictor might miss domains that the other captures. If you are more interested in manually curated domains corresponding to structural domains, you might rely more on the domain predictions manually validated in UniProtKB/Swiss-Prot. Thank you for contacting us, and I hope my answer helps you.