Annotator falsely returns Likely Oncogenic for a splice region variant without HGVSp info

[ibrahimkurt]

Hi,

We use vcf2maf w/ -inhibit-vep. Our maf file has splice region variants (naturally) without any HGVSp or HGVSp_Short info. When we call the mafannotator.py script with -q HGVSp or -q HGVSp_Short; the variants are falsely returned as Likely Oncogenic, although the variants do not exist in the OncoKB database. It even prints out a PubMed reference claimed to be related to the variant (https://pubmed.ncbi.nlm.nih.gov/21896780/). But it is not. The script works correctly though if we use -q Genomic_Change flag. We believe there could be a serious bug here for such variants.

The line of code:

${mafannotator} -i ${maffile} -o ${outputfile} -t NSCLC -q Genomic_Change -r GRCh38 -b ${token}

And here is the MAF file (changed to .txt to be able to upload to GitHub) showing only one of the problematic variants in it:

sample.txt

Intersting thing is, if you try URL/api manually like the following for the variant, it works correctly:

curl -X GET "https://www.oncokb.org/api/v1/annotate/mutations/byGenomicChange?genomicLocation=17,12107914,12107914,G,T&tumorType=NSCLC&referenceGenome=GRCh38" -H "accept: application/json" -H "Authorization: Bearer ***"

To understand where the bug might be, we added a print function to get the actual url submitted to OncoKB by the AnnotatorCore.pyscript. But for some reason it never prints out anything for any one of the -q parameter.

def pull_genomic_change_info(queries, annotate_hotspot):
    url = oncokb_annotation_api_url + '/annotate/mutations/byGenomicChange'
    response = makeoncokbpostrequest(url, queries)
    if response.status_code == 401:
        raise Exception('unauthorized')
    annotation = []
    if response.status_code == 200:
        annotation = response.json()
    else:
        for query in queries:
            geturl = url + '?'
            geturl += 'genomicLocation=' + query.genomicLocation
            geturl += '&tumorType=' + query.tumorType
            print(geturl)
            getresponse = makeoncokbgetrequest(geturl)
            if getresponse.status_code == 200:
                annotation.append(getresponse.json())
            else:
                # if the api call fails, we should still push a None into the list
                # to keep the same length of the queries
                print('Error on annotating the url ' + geturl)
                annotation.append(None)

    processed_annotation = []
    for query_annotation in annotation:
        processed_annotation.append(process_oncokb_annotation(query_annotation, annotate_hotspot))
    return processed_annotation

[ibrahimkurt]

A related but different serious problem:

A splice-site in intron, a LOW effect variant: PTCH1 c.3306+5G>T

It is returned as Likely Oncogenic along with FDA-approved drug recommendation with even -q Genomic_Change flag.

Interestingly, even manual URL/API as shown below returns the same wrong/false information:

curl -X GET "https://www.oncokb.org/api/v1/annotate/mutations/byGenomicChange?genomicLocation=9,95456271,95456271,C,A&tumorType=BREAST&referenceGenome=GRCh38" -H "accept: application/json" -H "Authorization: Bearer ***"

The really problematic thing is, the OncoKB webpage shows the correct information as unknown: https://www.oncokb.org/gene/PTCH1/c.3306+5G%3ET

We believe this shows that both the API and python-based annotator scripts could be problematic?

[ibrahimkurt]

Even a missense mutation with -q Genomic_Change gives the wrong output as Likely Oncogenic and conflicts with the webpage:

BRAF c.1633C>T L545F

https://www.oncokb.org/gene/BRAF/c.1633

[zhx828]

@ibrahimkurt few points to your questions

• we do not interprate c. through our website
• we look at consequence and variant classification when providing annotation. Even though you haven't provided any protein change, your file indicates this is splice variant therefor it will be mapped to our Truncating Mutations. It's likely oncogenic.
• 9,95456271,95456271,C,A is interpretated as X1102_splice by GenomeNexus, therefore is mapped to Level 3A Sonidegib, Vismodegib for Truncating Mutations
• please don't include your token in any public repo for security reason. I removed all from this post.

I hope these address your concerns.

We take annotation quality seriously. Happy to look into other variants if you think they have been interperted incorrectly.

[ibrahimkurt]

Thanks for the replies. I will in this case close this issue and open another one for HGVSp vs HGVSP_Short inconsistencies.