Reviewed as part of projects/NONPHYSIOLOGICAL_REACTIONS.md, specifically the
category-4 counter-example set: enzymes whose substrates are man-made but whose
functions are real and selected.
Atrazine is a synthetic herbicide introduced in the 1950s. A naive "synthetic
substrate implies non-physiological activity" filter would flag atzA. It should
not: Pseudomonas sp. ADP uses atrazine as a nitrogen source and atzA is the
committed first step PMID:8759853.
GO:0016810 (hydrolase acting on C-N but not peptide bonds) comes from
IPR011059, a metal-dependent hydrolase composite domain shared across the
amidohydrolase superfamily, whose best-known members are deaminases. AtzA
cleaves a C-halide bond and sits under GO:0019120, a sibling branch. REMOVE.
The instructive part: the 98%-identical paralog TriA (melamine deaminase),
differing at nine residues, does belong in GO:0016810
PMID:11274097.
So the term is right for the family and wrong for this member, which is exactly
what a fold-based inference cannot distinguish. Same failure mode as the
GO:0004301 epoxide hydrolase annotation on XANAU/dhlA.
GO:0016787 (root hydrolase) is redundant with the specific activity:
MARK_AS_OVER_ANNOTATED.
The Rhea-derived IEA here (GO_REF:0000120, RHEA:11312 + EC:3.8.1.8) lands on
the correct specific term GO:0018788 because RHEA:11312 is an exactMatch. That
is the arrangement RHEA:25968 should have on GO:0004090 rather than a
narrowMatch on the grouping class GO:0016616.
Which of AtzA and TriA is ancestral is unresolved, and no natural triazine
substrate has been tested for either.