Methodology Notes¶
The following methodology notes are documented for transparency.
Full detail is in DESIGN_PROVENANCE.md and in the
methodology_notes field of validation/all_predictions_summary.json.
Note 1 - Rosetta ΔΔG for D8PEA4 / D7BKC8 (cross-protein invalid)¶
Note 4 in DESIGN_PROVENANCE.md.
D8PEA4 and D7BKC8 are natural IS110 orthologs included in Strategy D.
Their Rosetta cartesian_ddg values were computed using IS621 as the
reference structure - this produces cross-protein absolute energies, not
thermodynamic ΔΔG values. Values set to None; ddg_method =
rosetta_cross_protein_INVALID.
Note 2 - Rosetta stability gate universally non-functional¶
Note 5 in DESIGN_PROVENANCE.md.
ALL 47 Rosetta cartesian_ddg values (Strategy A: 15, C: 2, D: 30) are
cross-structure absolute energies (range -31,838 to -41,308 kcal/mol),
not point-mutation ΔΔG values. The gate was not applied to any design.
Structural quality is instead proxied by ESMFold pLDDT (global mean > 70;
active-site > 70 for all triaged designs).
Disclosure (mandatory): “The Rosetta stability gate was non-functional for all designs due to cross-protein absolute energy computation. No designs were excluded on stability grounds. ESMFold pLDDT is used as a structural quality proxy.”
Note 3 - MHCflurry version shift¶
Strategy D S_Immuno scores individually recomputed with MHCflurry 2.2.1. Strategy B uses the IS621 conservative baseline (0.7594) rather than per-design computation. The calibrated IS621 lockpoint (0.9255) is reported as a secondary analysis alongside the verbatim 0.929.
Note 4 - P3 reference corrected¶
The earlier placeholder WT_IS621_S_IMMUNO = 0.250 is incorrect.
The correct value is 0.7594.
Delta for IS621_deimmunized_v2 = +0.1183 >= 0.10 -> PASS.
Note 5 - P5 diversity enforcement¶
Rank-5 was diversity-enforced: A_007 (pen_score = 0.9209, Strategy A) replaces the natural rank-5 D023 (pen_score = 0.9319, Strategy D) to satisfy the >= 3 strategy criterion. The top-5 is not purely rank-ordered.
Note 6 - Gate 8 ATLAS novelty not evaluated for Strategy B¶
The GENOME-ATLAS DuckDB embedding-distance gate was not evaluated in this
run (database access pending). Strategy B designs were pre-filtered for genus
diversity (700 distinct genera) at the sourcing step, providing de facto
novelty evidence. Disclosed in validation/P4_orthologs_result.json.