Antibody Mutation Recommendation Validator

What It Solves

Antibody mutation recommendations often fail not because a model cannot choose a site, but because CDR boundaries, 1-based vs 0-based indices, and claimed vs actual residues are mixed up. In the H-GAMMA-1 case, F35H, D26H, and N52H were all invalid: position 35 already contained His, while the other positions did not match their claimed amino acids.

How It Works

The skill turns that incident review into a validation checklist:
- Boundary validation: avoid hard-coded CDR ranges; prefer IMGT, Kabat, or conservative anchor motifs.
- Unified indexing: use one coordinate convention end to end, avoiding mixing Python 0-based and biology 1-based notation.
- Reverse validation: read back the actual residue at each recommended site and reject A→A or impossible substitutions.
- Context logging: print 3–5 neighboring residues and record parameters and versions for reproducibility.

Boundaries

It targets antibody mutation recommendations, low-pH stability engineering, and site sanity checks. The same reverse-validation principle can extend to genome coordinate mapping, ChIP-seq annotation, and gRNA design, but the exact numbering system, anchors, and database version still need sequence-specific confirmation.