Update `feature_truncation` in Nallo rank model?

[fellen31]

Should #88 be updated in the Nallo rank models as well? ping @ramprasadn @jemten

[jemten]

Sorry for the very late reply. Ve base the score on the severity as ranked by VEP
https://www.ensembl.org/info/genome/variation/prediction/predicted_data.html

The items on this list and the ranking of them can change between VEP versions and we try to check whenever we update VEP to see that we are somewhat right. In the case of #88 I think feature _truncation used to be much, much further down the list and that the PR simply brought it up to reflect the current VEP ranking.

[fellen31]

Okay. So looking at VEP 110 and the 1.39 short-read rank model:

• All HIGH impact consequences have score 8
  • except transcript_ablation which has score 10?
  • except transcript_amplification which has score 5?
  • except feature_elongation which has score 1?
• All MODERATE have score 5
• All LOW have score 5 (why same as MODERATE?)
  • except start_retained_variant which has score 1
  • except stop_retained_variant which has score 1
  • except synonymous_variant which has score 2 (?)
• All MODIFIER has score 1
  • except non_coding_transcript_exon_variant which has score 3 (?)
  • except intergenic_variant which has score 0 (?)

I see no difference between VEP 110 and 115.

[fellen31]

Okay. So looking at VEP 110 and the 1.39 short-read rank model:

• All HIGH impact consequences have score 8

  • except transcript_ablation which has score 10?
  • except transcript_amplification which has score 5?
  • except feature_elongation which has score 1?

• All MODERATE have score 5

• All LOW have score 5 (why same as MODERATE?)

  • except start_retained_variant which has score 1
  • except stop_retained_variant which has score 1
  • except synonymous_variant which has score 2 (?)

• All MODIFIER has score 1

  • except non_coding_transcript_exon_variant which has score 3 (?)
  • except intergenic_variant which has score 0 (?)

I see no difference between VEP 110 and 115.

I'm also wondering about the SV model using

  [[coding_sequence_variant]]
    score = 7
    priority = 5
    string = 'coding_sequence_variant'

while the SNV model is using

  [[coding_sequence_variant]]
    score = 1
    priority = 2
    string = 'coding_sequence_variant'

[fellen31]

I tried updating feature_truncation, but it results in an almost 3x increase of SVs uploaded to Scout. 439 of variants uploaded to Scout with updated score for feature_truncation were feature_truncation&intron_variant deletions. Until there's a way to give separate scores for feature_truncation and feature_truncation&intron_variant, I think I can't change this.

It should be noted though that by not updating the score we might miss some interesting feature_truncations: 23 variants in a NIST case were feature_truncation but not intron_variant. 10 of these got scores high enough to be uploaded to Scout. Were we to keep feature_truncation at the current score, this results in 5 variants not being uploaded to Scout (three 3_prime_UTR_variant, one frameshift_variant, and one stop_lost).