Skip to content

Experimental window exposure

Explore delivery assumptions with the MHCtools exposure model. This analysis leaves vaccine ranking and products unchanged.

Run

Start with an enriched --output-epitopes file containing native antigen contexts and saved per-allele scores. The command evaluates all admitted, target-containing SLP windows in the length range, before ranking or count limits.

Save this minimal profile.json; kinetics remain unassessed until supplied:

{
  "scenario": "IM SLP comparison; kinetics not supplied",
  "times_hours": [0, 1, 2, 4, 8, 24, 48],
  "administration": "IM"
}
vaxrank-window-exposure --input-epitopes native.tsv --profile profile.json \
  --min-length 20 --max-length 30 --n-term free --c-term free \
  --output-dir window-exposure

Outputs: request.json, exposure.json, cores.csv and report.md. The request also works with mhctools vaccine-windows.

Alleles in saved scores and supplied core_intervals use the same Topiary/mhcgnomes validation as prediction inputs. Aliases match their canonical names, including nonhuman alleles and class II pairs. Genes, serotypes, allele groups and null alleles are rejected instead of being treated as expressed prediction alleles. The library's normalize_mhc_allele follows the same rule.

Interpret

  • Free linear peptides default to 1 µg per window. Supply the actual terminal chemistry; modified peptides remain unassessed by the current engine.
  • Missing context, scores or admission appear as coverage gaps. No new prediction or source reconstruction occurs; provenance and admission records are retained.
  • Supply explicit kinetic rates and cleavage assessments under kinetics. Per-window overrides use candidates[window_id].
  • The tracked interval is the saved scored ligand. A shorter documented core uses core_intervals with prediction ID, allele, source and zero-based, half-open coordinates within that ligand. Affinity alone does not identify a shorter core. Original ligand scores and provenance remain recorded.
  • APC arrival, blood persistence and pMHC exposure are different outputs. Core marginals cannot be summed as unique dose fractions; weighted means require numerical assessments for every contributing core.

For rates, units and sensitivity examples, see the MHCtools window guide. For mRNA, use the full-construct model with the actual encoded construct, transcript dose and delivery assumptions. The SLP command does not infer an RNA sequence from a protein window.