GitHub
09/21/2026, 6:46 AMtestdata/variants/variants_chr4_chr19_germline.vcf — HaplotypeCaller-style, one NORMAL sample, two records:
• 4:1801844 C>G — FGFR3 p.Pro250Arg, the Muenke syndrome mutation. Three bases from the somatic FGFR3 S249C in variants_chr4_chr19.vcf, so pvacseq folds it into that window. It is applied to the wild-type window as well as the mutant one, which is the point: the wild-type sequence becomes patient-specific rather than reference.
• 19:35878587 C>T — APLP1 p.Ser528Phe, 565 bases from the nearest somatic call. Outside every window, so it must never appear in the peptide output. This is the control that germline calls stay context and never become candidates.
• testdata/sample_sheets/sample_sheet_variants_grch38_germline.csv — pairs the existing somatic VCF with the germline one through a germline_vcf column.
Verified
Both variants were annotated against the chr4/chr19 VEP cache subset already in this branch, and the effect was confirmed by running pvacseq generate_protein_fasta with and without the germline context. The two codons of interest are adjacent, 249 carrying the somatic change and 250 the germline one:
249 250
no germline WT DVLER S P HRPI reference at both codons
MT DVLER C P HRPI somatic S249C only
with germline WT DVLER S R HRPI germline P250R only
MT DVLER C R HRPI both changes
Read down a column for the somatic change (S vs C) and across the pairs for the germline one (P vs R). The germline variant reaches the wild-type window too, which is what makes the wild-type sequence the patient's own. The distant APLP1 variant is dropped before annotation.
The consuming pipeline change is in progress on nf-core/epitopeprediction#362.
🤖 Generated with Claude Code
nf-core/test-datasetsGitHub
09/21/2026, 7:12 AMGitHub
09/21/2026, 7:12 AM