A case of a missing VFR_ED_TR in precision.vcf

[bricoletc]

Hey Marto,

I have the following case from a cortex-evaluated vcf (organism: Pf, gene: EBA175) (vcf fields broken into lines on purpose)

Pf3D7_07_v3
1360357
UNION_BC_k31_var_11058
GCACTGAAATAGCACACAGAACGGAAACTCGTACGGATGAACGAAAAAATCAGGAACCAGCAAATAAGGATTTAAAGAATCCACAACAAAGTGTAGGAGAGAACGGAACTAAAGATTTATTACAAGAAGATTTAGGAGGATCACGAAGTGAAGACGAAGTGACACAAGAATTTGGAGTAAATCATGGAATACCTAAGGGTGAGGATCAAACGTTAGGAAAATCTGACGCCATTCCAAACATAGGCGAACCCGAAACGGGAATTTCCACTACAGAAGAAAGTAGACATGAAGAAGGCCACAATAAACAAGCATTGTCTACTTCAGTCGATGAGCCTGAATTATCTGATACACTTCAATTGCATGAAGATACTAAAGAAAATGATAAACTACCCCTAGAATCATCTACAATCACATCTCCTACGGAAAGTGGAAGTTCTGATACAGA
ACACTGAAATAGCACACAGAAC
.
PASS
KMER=31;SVLEN=-423;SVTYPE=COMPLEX
GT:COV:GT_CONF:VFR_ED_RA:VFR_ED_TA:VFR_ALLELE_LEN:VFR_ALLELE_MATCH_COUNT:VFR_ALLELE_MATCH_FRAC:VFR_IN_MASK:VFR_RESULT
1/1:0,33:235.46:2:0:22:22:1.0:0:TP

The formula for precision is essentially (VFR_ED_TR - VFR_ED_TA )/ VFR_ED_TR, but VFR_ED_TR is missing. Though if we reason here, VFR_ED_TA is 0, so cortex's call is perfect, so VFR_ED_TR = edit_distance(ref,alt).

i)Is that reasoning at the end correct?
ii)Is this a bug so to speak. If so I can probably push a PR in doing the above if you want

[martinghunt]

i) I think the reasoning is correct.
ii) It's a "feature". See the comments:

varifier/varifier/vcf_stats.py

Lines 91 to 101 in a6938f9

	# Can have cases where VFR_ED_TR is not present. This is the edit
	# distance between the truth and ref allele. Calculated from mapping
	# of ref probe to the truth.
	# Cases are:
	# - FP because the alt probe does not map to the truth ref. In this
	# case we don't map the ref probe to the truth.
	# - FP, alt probe mapped, but the truth probe did not map
	# - TP or Partial_TP, and we have VFR_ED_RA,VFR_ED_TA but not VFR_ED_TR.
	# This is where alt probe mapped, but ref probe did not map, or no
	# mapping found in same place as the alt mapping
	# (either way, is counted as no hit).

I don't mind if you want to change the behaviour to add in VFR_ED_TR, but I'm not sure about edge cases that we might not think of. Can imagine a boatload of tests failing if you try changing, but go for it if you want...