HGVSc/HGVS_OFFSET emitted for non-coding indels where VEP produces empty

[mwiewior]

Summary

vepyr computes HGVSc and HGVS_OFFSET for certain non-coding transcript indels (n. notation) and 3'UTR indels (c.* notation) where Ensembl VEP leaves both fields empty. This causes 35 HGVSc + 35 HGVS_OFFSET = 70 field-level mismatches on chr6 of the GIAB HG002 benchmark (271,966 variants, 2,371,184 CSQ entries).

Pattern

All 35 mismatches share the same structure:

• vepyr produces a valid HGVSc (e.g., ENST00000663428.2:n.1676_1677del) and a non-zero HGVS_OFFSET
• VEP produces empty string for both HGVSc and HGVS_OFFSET

The affected transcripts are either:

1. Non-coding transcripts — HGVSc uses n. prefix (e.g., lncRNAs, processed pseudogenes)
2. 3'UTR indels — HGVSc uses c.* prefix (downstream of stop codon)

All are indels in repeat contexts that require 3' HGVS shifting (hence the non-zero HGVS_OFFSET).

Hypothesis

VEP may skip HGVSc computation when the 3' shift lands outside the transcript's spliced sequence, or when _return_3prime() fails for non-coding transcripts that lack a spliced_seq. In these cases VEP silently returns undef and emits empty, while vepyr's genomic shift path succeeds and emits a value.

It's also possible this is a vepyr improvement — the values appear correct, and VEP's omission may be a bug. Needs investigation of VEP's _calc_hgvs() for non-coding transcripts to determine whether the empty output is intentional.

Examples

Non-coding transcripts (n. notation)

Variant	Transcript	vepyr HGVSc	vepyr HGVS_OFFSET	VEP
chr6:2307503 CTA>C	ENST00000663428.2	n.1676_1677del	25	empty
chr6:2307503 CTA>CTATA	ENST00000663428.2	n.1677_1678insAT	25	empty
chr6:4490858 AAAG>A	ENST00000827273.1	n.1058_1060del	4	empty
chr6:6719895 TAAA>T	ENST00000776254.1	n.1051_1053del	14	empty
chr6:8441367 CAAA>C	ENST00000790778.1	n.783_785del	15	empty
chr6:10456777 C>CAA	ENST00000366312.2	n.666_667insAA	14	empty
chr6:22194385 CAA>C	ENST00000606851.5	n.1901_1902del	13	empty
chr6:22194385 CAA>C	ENST00000822831.1	n.664_665del	13	empty
chr6:22194385 CAA>C	ENST00000822928.1	n.720_721del	13	empty

3'UTR (c.* notation)

Variant	Transcript	vepyr HGVSc	vepyr HGVS_OFFSET	VEP
chr6:25781174 CAGAA>C	ENST00000377905.9	c.*2012_*2015del	25	empty

Impact

• 35 / 2,371,184 CSQ entries (0.001%)
• vepyr produces more data than VEP, not wrong data
• All emitted values appear structurally valid (correct transcript ID, correct notation prefix, plausible positions)

Investigation needed

1. Check VEP Perl _calc_hgvs() / _return_3prime() for non-coding transcripts — does it intentionally skip, or fail silently?
2. Verify the emitted positions are numerically correct by manual cDNA coordinate lookup
3. Decide: match VEP (suppress) or accept as improvement

Context

Identified in vepyr chr6 fast annotation benchmark after fixing #99 (HGVS_OFFSET flank buffer). These mismatches are unrelated to #99 — they exist because vepyr computes HGVS for transcripts where VEP doesn't, not because of incorrect shift values.