most severe consequence polypyrimidine stretch

[dnil]

Describe the bug

We are seeing more issues with the most severe consequence. We recently fixed the clinical filter in scout (Clinical-Genomics/scout#3858) to allow showing these nice little fellows, but that falls a bit short when the variant is not loaded. Currently, they also get a penalty from the rank model for being less clinically important, like this causative one from pureamoeba:

1       216210952       SV_221_1:tiddit|CNVnator_del_142:cnvnator|MantaDEL:10115:0:1:0:0:0:manta        N       <DEL>   .       PASS    Annotation=USH2A;CIEND=0,2;CIPOS=0,2;CSQ=deletion|splice_polypyrimidine_tract_variant&coding_sequence_variant&intron_variant&feature_truncation|LOW|USH2A|7399|Transcript|NM_206933.4|protein_coding|22-32/72|21-32/71||||||||||-1||EntrezGene|12601|YES||||NP_996816.3||||||RefSeq|||||||;CTG=CTCTTAGTAGCCATTGCTTTGATTGTTCTTTGTTTTCAGGATTGTACTGGTAAAGCCATATTTCAACTCCTATTACAATTATTTGATAAAATGTTTCAGGATATTGATTCCACCTGTTTAAAATTTCCATTGAAATTTGTGAGCCTAAAAATCAATGAGTGAAGAGTAAGTTTCAAAGGATTCAAATTAAATGCTGTTGTAGCAAAACTGTCAATTATCCCGCATTCTTTTATTAATTTTGCTTTTATTTGTAGAATGGCCTTTCCT;END=216301040;FOUNDBY=3;GeneticModels=pureamoeba:AD;HOMLEN=2;HOMSEQ=AA;IMPRECISE;LFA=16,11;LFB=19,11;LTE=14,11;REGIONA=216210823,216210952;REGIONB=216301040,216301211;RankResult=4|-12|5|0|3|0|3|0;RankScore=pureamoeba:3;SUPP_VEC=111;SVLEN=90088;SVTYPE=DEL;VARID=CNVnator_del_142:cnvnator|MantaDEL:10115:0:1:0:0:0:manta;clingen_cgh_benign=1;clinical_genomics_loqusFrq=0.00042;clinical_genomics_loqusObs=3;cnvnator_CHROM=CNVnator_del_142|1;cnvnator_FILTERS=CNVnator_del_142|PASS;cnvnator_INFO=CNVnator_del_142|END:216301000|SVTYPE:DEL|SVLEN:-90000|IMPRECISE|natorRD:0.498228|natorP1:1.77081e-12|natorP2:0|natorP3:1.81105e-12|natorP4:0|natorQ0:0.0024426;cnvnator_POS=CNVnator_del_142|216211001;cnvnator_QUAL=CNVnator_del_142|.;cnvnator_SAMPLE=CNVnator_del_142;manta_CHROM=MantaDEL_10115_0_1_0_0_0|1;manta_FILTERS=MantaDEL_10115_0_1_0_0_0|PASS;manta_INFO=MantaDEL_10115_0_1_0_0_0|END:216301039|SVTYPE:DEL|SVLEN:-90090|CIPOS:0:2|CIEND:0:2|HOMLEN:2|HOMSEQ:AA;manta_POS=MantaDEL_10115_0_1_0_0_0|216210949;manta_QUAL=MantaDEL_10115_0_1_0_0_0|819;manta_SAMPLE=MantaDEL_10115_0_1_0_0_0;most_severe_consequence=12601:deletion|splice_polypyrimidine_tract_variant;natorP1=1.77081e-12;natorP2=0;natorP3=1.81105e-12;natorP4=0;natorQ0=0.0024426;natorRD=0.498228;pureamoeba_tiddit_ACC11683A12_CHROM=SV_221_1|1;pureamoeba_tiddit_ACC11683A12_FILTERS=SV_221_1|PASS;pureamoeba_tiddit_ACC11683A12_INFO=SV_221_1|SVTYPE:DEL|SVLEN:90088|END:216301040|REGIONA:216210823:216210952|REGIONB:216301040:216301211|LFA:16:11|LFB:19:11|LTE:14:11|CTG:CTCTTAGTAGCCATTGCTTTGATTGTTCTTTGTTTTCAGGATTGTACTGGTAAAGCCATATTTCAACTCCTATTACAATTATTTGATAAAATGTTTCAGGATATTGATTCCACCTGTTTAAAATTTCCATTGAAATTTGTGAGCCTAAAAATCAATGAGTGAAGAGTAAGTTTCAAAGGATTCAAATTAAATGCTGTTGTAGCAAAACTGTCAATTATCCCGCATTCTTTTATTAATTTTGCTTTTATTTGTAGAATGGCCTTTCCT;pureamoeba_tiddit_ACC11683A12_POS=SV_221_1|216210952;pureamoeba_tiddit_ACC11683A12_QUAL=SV_221_1|.;pureamoeba_tiddit_ACC11683A12_SAMPLE=SV_221_1|ACC11683A12|GT:0/1|CN:1|COV:37:16.750621529891426:36|DV:14|RV:11|LQ:0.0:0.0|RR:22:20|DR:24:30;set=Intersection;svdb_origin=tiddit|cnvnator|manta;tiddit_CHROM=SV_221_1|1;tiddit_FILTERS=SV_221_1|PASS;tiddit_INFO=SV_221_1|SVTYPE:DEL|SVLEN:90088|END:216301040|REGIONA:216210823:216210952|REGIONB:216301040:216301211|LFA:16:11|LFB:19:11|LTE:14:11|CTG:CTCTTAGTAGCCATTGCTTTGATTGTTCTTTGTTTTCAGGATTGTACTGGTAAAGCCATATTTCAACTCCTATTACAATTATTTGATAAAATGTTTCAGGATATTGATTCCACCTGTTTAAAATTTCCATTGAAATTTGTGAGCCTAAAAATCAATGAGTGAAGAGTAAGTTTCAAAGGATTCAAATTAAATGCTGTTGTAGCAAAACTGTCAATTATCCCGCATTCTTTTATTAATTTTGCTTTTATTTGTAGAATGGCCTTTCCT|SUPP_VEC:1;tiddit_POS=SV_221_1|216210952;tiddit_QUAL=SV_221_1|.;tiddit_SAMPLE=SV_221_1 GT:CN:COV:DV:RV:LQ:RR:DR        0/1:1:37,16.7506,36:14:11:0,0:22,20:24,30

It's a clear het deletion of several exons, called by all callers, but the most severe consequence becomes splice_polypyrimidine_tract_variant.

We can update the SV rank score model to also include this as a (potentially) severe consequence for now perhaps?

Any chances of getting VEP adapt, eg the SO scale? I still feel from the textual definition, these warrant a "transcript_ablation" though the entire transcript is not gone.

Other options could include to revert to having coding_sequence_variant higher on the severity tree. It seems to be a fallback that is usually present on this kind of variants.

Software version (please complete the following information):

• MIP: 11.1.3

Additional context
Add any other context about the problem here.

[jemten]

Thanks for reporting @dnil,
I believe that what's mainly driving this variant down in the ranking is that it is flagged as annotated as benign according to clingen_cgh so it get's a penalty of 12 points. Need to investigate further how that annotation ended up there. From a consequence perspective it gets 5 points from being a splice_polypyrimidine_tract_variant.

Can you clarify a little on what you mean with: Any chances of getting VEP adapt, eg the SO scale?

[dnil]

Ooh, nice catch - I didn't check what field had the -12! It is a recessive condition with some carriers, so not completely strange to find it among variants found benign on array at some point. I can talk to the array ppl about that part!

As for the VEP part, I mean even if this would not have got the 10 I feel it should have for transcript ablation, it would at the very least get its 7 for coding_sequence_variant if that was higher prio for most_severe_consequence than the lower scoring (5) splice_polypyrimidine_tract_variant.

We could patch that on our with raising the latter to score 7; which might arguably be consistent.

Or by swapping the order for most severe consequence prio, but then it would feel better to also discuss a bit with the VEP/SO folks about it.

[dnil]

I've contacted the aCGH folks, but if we can't find a simple solution for the db-export to properly cover the zygosity of calls, just ditching the benign aCGH track seems like a good option. We were about to do that for the hg38 migration anyway.

It would have helped, but fiddling with a two point difference in score for functional annotation seems moot when we have that huge malus to axe.

Since we have a fairly big normal db anyway, we should make do just fine without the aCGH-benign track, possibly leaning towards increasing the penalty for being common locally a bit to compensate the missing benigns then, if a verification is still called for anyway.

[jemten]

As for the VEP part, I mean even if this would not have got the 10 I feel it should have for transcript ablation, it would at the very least get its 7 for coding_sequence_variant if that was higher prio for most_severe_consequence than the lower scoring (5) splice_polypyrimidine_tract_variant.

Hmm for the structural variants we do rank coding_sequence_variant as a more severe consequence than the splice_polymeridine_tract one so it should have gotten a 7. I need to look into whats going on here.

Let me know if you think we should remove the benign cgh database. But yes this would mean a new validation and thus we would probably do it when we update tiddit and update our loqusDB dump.

[dnil]

I am leaning that way: this would still be an issue with several recessive disorders with somewhat recurrent het carriers. Agne had some idea about another recurrent CNV file that he and Jesper made - waiting for a confirmation on wether these are still in there.

[dnil]

Had another discussion with Agne: it was the CNV cluster track they used for collapsing the local aCGH db a bit, and weeded to not include pathogenic calls from said db. The same issue kind of remains, and this is a much smaller set than say loqus or svdb. I think we are still better off with the local data for count/frequency info, and then the pathogenic calls are great to complement the lack of SVs matching from ClinVar.

[jemten]

Right, so to summarise you would like to not include the benign variants from the cgh array in the rankscore. However, we will still annotate called SV:s using that file. Did I get this right?

[dnil]

Yes!

On a lower prio, there is still something a little funny about the sort order for so term severity and the rank model scores not lining up - can't help but think that can come back and bite, but it would have felt best to correct at the source.