When gene_kos are used alongside reaction_kos and reaction_kis in compute_strain_designs, some returned interventions fail to satisfy the PROTECT module — e.g., a simple BIOMASS >= min_growth constraint that the protected target is required to support.
This has not been observed when only reaction_kis / reaction_kos are used, which suggests the issue lies in how gene-level interventions are translated through GPR rules and reflected back into the reaction-level design space.
Expected: every returned design satisfies all PROTECT modules.
Observed: a subset of returned designs violates the PROTECT constraint when re-evaluated on the perturbed model.
Suspected cause: compression may be lumping knock-in reactions with knock-out reactions in a way that interacts badly with the new GPR rule integration. If true, the base MILP is fine and the bug lives in the compression / GPR-extension boundary. The two recent large changes (compression rework + GPR propagation through compression) are the prime suspects.
Possibly related: #43 (gene-level designs containing gene KOs with no reaction-level effect) — same intervention-space mapping concern.
Reproduction: TBD.
When
gene_kosare used alongsidereaction_kosandreaction_kisincompute_strain_designs, some returned interventions fail to satisfy thePROTECTmodule — e.g., a simpleBIOMASS >= min_growthconstraint that the protected target is required to support.This has not been observed when only
reaction_kis/reaction_kosare used, which suggests the issue lies in how gene-level interventions are translated through GPR rules and reflected back into the reaction-level design space.Expected: every returned design satisfies all PROTECT modules.
Observed: a subset of returned designs violates the PROTECT constraint when re-evaluated on the perturbed model.
Suspected cause: compression may be lumping knock-in reactions with knock-out reactions in a way that interacts badly with the new GPR rule integration. If true, the base MILP is fine and the bug lives in the compression / GPR-extension boundary. The two recent large changes (compression rework + GPR propagation through compression) are the prime suspects.
Possibly related: #43 (gene-level designs containing gene KOs with no reaction-level effect) — same intervention-space mapping concern.
Reproduction: TBD.