/MAGE

<!--

COPYRIGHT NOTICE

This file is part of the "Universal Biomedical Skills" project.

Copyright (c) 2026 MD BABU MIA, PhD <md.babu.mia@mssm.edu>

All Rights Reserved.

#

This code is proprietary and confidential.

Unauthorized copying of this file, via any medium is strictly prohibited.

#

Provenance: Authenticated by MD BABU MIA

-->

--- name: mage-antibody-generator description: Ab seq forge keywords:

• antibody
• antigen
• FASTA
• generation
• validation

measurable_outcome: Generate the requested number of antibody sequences (default ≥5) with metadata (model checkpoint, seed) and deliver FASTA files within 10 minutes. license: MIT metadata: author: MAGE Team version: "1.0.0" compatibility:

• system: Python 3.9+ / GPU

allowed-tools:

• run_shell_command
• read_file

---

MAGE (Monoclonal Antibody Generator)

Run the MAGE antibody generation workflow to propose antigen-conditioned antibody sequences for downstream structural validation.

Workflow

1. **Prep env:** `cd repo` and install dependencies, then point to GPU if available. 2. **Run generator:** `python generate_antibodies.py --antigen_sequence <SEQ> --num_candidates N --output_dir ./results`. 3. **Collect outputs:** Provide FASTA paths + metadata, optionally translate into JSON manifest. 4. **Recommend validation:** Suggest AlphaFold/Rosetta checks and wet-lab follow-up.

Guardrails

• Never imply binding efficacy without structural/experimental confirmation.
• Track model version + seeds to ensure reproducibility.
• Encourage downstream filtering (liability motifs, developability metrics).

References

• Source instructions in `README.md` and repo scripts.

<!-- AUTHOR_SIGNATURE: 9a7f3c2e-MD-BABU-MIA-2026-MSSM-SECURE -->