# Rank SOD1 variants by dimerization

453 single-amino-acid variants of human SOD1 were measured in yeast in three biological replicates. Predict their heterodimerization with wild-type SOD1, using binding protein-fragment complementation (bPCA). Variant SOD1 and wild-type SOD1 are fused to complementary DHFR fragments; their association allows growth under methotrexate selection. The readout depends on both folded abundance and association, and is not a purified binding affinity or an abundance-adjusted residual. Rank the supplied variants by the normalized aggregate experimental score, highest first. Normalization places wild-type-like abundance or heterodimerization near 0 and strongly disrupted variants near -1; values above 0 are possible.

Each supplied sequence is full-length human SOD1, including its initial methionine. The common wild-type sequence is:
MATKAVCVLKGDGPVQGIINFEQKESNGPVKVWGSIKGLTEGLHGFHVHEFGDNTAGCTSAGPHFNPLSRKHGGPKDEERHVGDLGNVTADKDGVADVSIEDSVISLSGDHCIIGRTLVVHEKADDLGKGGNEESTKTGNAGSRLACGVIGIAQ

Return exactly one JSON object: {"ranking": ["P001", "P002", "..."]}. Include every supplied candidate ID exactly once, highest predicted assay score first. No additional fields or explanatory text. A single JSON code fence is accepted.

You receive the assay description and candidate sequences. No tools, code execution, internet, external retrieval, or experimental training labels are available. The harness writes your response to /app/answer.json.
