ImmunoSeq
Compare antibody self-library coverage.
Results
Loading starting example…
Method & evidence
Score variable-region sequences with exact 8–12-mer human-proteome, paired-human-antibody, and paired-mouse-antibody Bloom-filter libraries, then optionally rank explicitly bounded framework substitutions.
Scope & limitations
- The released bundle uses a frozen current OAS snapshot and is explicitly not paper-identical because the authors did not publish their 15 filters or frozen source corpus.
- Results from this snapshot must not be presented as exact reproduction of paper tables.
- All reference assets are checksum-pinned on Azure Files and remain outside container layers.
Interpreting results
- Higher hit rate means greater membership in these sequence libraries; it is not a calibrated clinical immunogenicity probability.
- The paired-mouse negative penalty is fixed at 0.2 to preserve the published objective.
- Optimization ranks the same computational objective and requires experimental and developability review.
References
Input requirements & outputs
Input checks
- Canonical amino acids only; 40–180 residues per chain
- Scoring accepts 1–32 rows
- Optimization accepts one row and at most 80 sorted, unique, one-based mutable positions
- Mutable positions must be framework positions selected and reviewed outside ImmunoSeq
Expected results
- Annotated score Dataset — Row-preserving global, chain, residue, reference, and result evidence.
- Ranked substitution Dataset — Mutation identity, complete candidate sequences, score delta, and candidate digest.
- Immutable native evidence — Exact request, source revision, reference manifest, method, row, candidate, and result digests.
