VEP output SIFT_score unclear

Question

We have been experimenting with VEP (Variant Effect Predictor). One of the meta data attributes that we are interested in is the SIFT score, indeed when we apply the dbNSFP plug we get a column containing the scores (named SIFT_score). However, I don't understand why there are sometimes dots or multiple values in the fields. For example, the gene ENSG00000196924 below has 5 transcripts:

The SIFT_score column contains several values, not 1 per transcript/rs-number...

Here is another example that confuses me (I added the SIFT_pred column this time):

There are two mutations, the lower one can be expressed in 4 transcripts and I thus understand that there can be 4 SIFT scores, but why are all for in given in every row? Is the first one the SIFT_score for the first transcript?

One last example, again 4 transcripts, but now 2 of the scores are dots, what does that mean?

I have been looking for quite some time now how to interpret this data, any help is appreciated.

Answer 1

The dbNSFP plugin from VEP accesses tables of data for each variant from dbNSFP and pulls out the values. dbNSFP provide their SIFT scores in that format: a score for every transcript affected by the variant, all on one line. The lookup is just for the variant, not for the variant/transcript combo, so they provide scores for all variant/transcript combos. You can also get a column that gives you a list of the transcripts or proteins (Ensembl_transcriptid or Ensembl_proteinid) in order so you know which score goes with which transcript.

A better way to get SIFT scores with VEP is to get them directly from VEP, rather than using dbNSFP. This will get the SIFT score that goes with the transcript on the line with the relevant transcript.

Answer 2

The first gene you mention, ENSG00000196924, actually has 6 transcripts (link to the VarSome.com page of variant rs371839875), not 5. It's just that one of them is non-coding:

So the Sift scores you see are indeed one per transcript, it's just that there are 6 because dbNSFP also includes a score for the non-coding transcript of the gene.

The dots are just there as placeholders, they mean there was no value associated with that transcript. Many tools will show some sort of symbol instead of an empty field both for clarity and for practical technical reasons.

Visiting the variant's page on VarSome gives you a clearer picture since we collapse the identical scores and also include the converted rankscore provided by dbNSFP so you can have a single number for your variant:

---

Disclaimer: I work for the company behind VarSome, but it's a free tool. You need to pay to annotate VCF files (unlike the 100% free VEP), but it's free to use as a lookup tool for single variants.