The selected horse ALDH7A1 A0A9L0RRL6 is521 aa versus canonical human 539 aa. Alignment identifies an internal deletion of human 65–82; the human structure has a beta strand at 65–70, so this should not be dismissed as a variable terminal peptide. Catalytic Glu296/Cys330 and the annotated nucleotide/substrate-binding positions are preserved; the N-terminal mitochondrial precursor segment is present. Broad ALDH chemistry and mitochondrial targeting remain reasonable inferences, but exact substrate kinetics and effects of the deletion are unresolved.
PMID:20207735(https://pubmed.ncbi.nlm.nih.gov/20207735/), DOI 10.1074/jbc.M109.077925, reports purified recombinant human enzyme metabolizing betaine aldehyde, lipid-peroxidation aldehydes and alpha-aminoadipic semialdehyde. PMID:16491085(https://pubmed.ncbi.nlm.nih.gov/16491085/) establishes pathogenic ALDH7A1 loss of alpha-AASA dehydrogenase activity. Neither inspected source provides a negative glyceraldehyde-3-phosphate assay. Broad ALDH substrate promiscuity cannot validate the ProtNLM substrate prediction, but it also makes a confident rejection based only on the canonical substrate unsafe.
The selected horse protein is not assumed to represent all human-like cytosolic isoforms. The horse GOA localization rows beyond mitochondria require isoform-specific evidence. A completed human Falcon report has been inspected; horse-specific Edison was not indicated by genome-only source references.
Conservation of the catalytic residues does not establish an intact active fold after the internal deletion. Broad aldehyde-dehydrogenase and oxidoreductase activity therefore remain UNDECIDED for this exact model as well as the substrate-specific terms; the ALDH7A1 family assignment itself remains well supported.
Deep research. just deep-research-falcon HORSE ALDH7A1 --fallback perplexity-lite failed (falcon: "Failed to authenticate"; perplexity-lite: HTTP 401 insufficient quota). No deep-research file was written. The review relies on cached publications (PMID:20207735, PMID:16491085, PMID:21338592, PMID:31492851, PMID:40233740, PMID:36225138), the human ALDH7A1 review, and new sequence/gene-model analysis.
Identity of A0A9L0RRL6 (alternative gene model, exon-skip isoform). New analysis ALDH7A1-bioinformatics/compare_horse_models_rereview.py → RESULTS-rereview.md / results-rereview.json (live UniProt + Ensembl). The horse ALDH7A1 gene ENSECAG00000010547 has five Ensembl protein-coding models; none is reviewed in UniProt.
- Ensembl canonical ENSECAT00000079051 → A0A9L0R3Y5, 566 aa, no internal gap vs human, with a 27-residue N-terminal extension upstream of the human-equivalent Met1.
- Benchmark A0A9L0RRL6 = ENSECAT00000119202, 521 aa (human P49419 is 539 aa). Its coding exons are [192, 66, 81, ...] versus [273, 54, 66, 81, ...] in the canonical transcript: it skips the 54-nt second coding exon. This is the 18-residue in-frame loss of human 65-82 ("VITTYCPANNEPIARVRQ"), covering beta strands 65-70 and 75-81 at the start of the PF00171 aldehyde-dehydrogenase domain. No human isoform has this skip (human isoforms 2 and 4 alter 1-28 and 337-400).
- Retained: start at the mitochondrial-precursor Met, with all 26 human transit-peptide positions paired; catalytic Glu (horse 278 = human 296) and Cys (horse 312 = human 330); all mapped NAD/substrate-binding residues; 95.6% identity over paired residues.
- Other models: A0A5F5PY19 has a 14-residue gap over human 314-327 that replaces the catalytic Cys330 with Arg. A0A9L0RA17 lacks human 291-304, including Glu296. F7CZV6 is C-terminally truncated (502 aa). K9KBK1 is a fragment.
Evolutionary grounding. UniProt places horse ALDH7A1 (all models) in PANTHER PTHR43521:SF1, the same subfamily as the mouse (Q9DBF1), rat (Q64057) and bovine (Q2KJC9) orthologs, and zebrafish aldh7a1. Human P49419 is classed in SF5, but PAINT node PTN000192435 is seeded by human, mouse (MGI:108186) and zebrafish, so it spans that SF1/SF5 split. It sits at or above the bony-vertebrate ancestor, and horse descends from it. PAINT asserted GO:0004043, GO:0008802, GO:0019477, GO:0005739, GO:0005829, GO:0005634, GO:0000139 and GO:0005886 at that node. GO:0004029 was asserted at the deeper PTN000192422 (seeds human ALDH7A1 + Arabidopsis AT5G62530, which is ALDH12A1/P5CDH). The plant-only node PTN000192427 asserts P5C dehydrogenase and proline catabolism; those terms are restricted to the ALDH12 branch and are deliberately not propagated to ALDH7A1. No horse product currently carries IBA rows. The canonical A0A9L0R3Y5 gets Ensembl Compara projections that the benchmark accession does not.
Decision on the isoform problem. The prior pass marked every function UNDECIDED because the exon skip could compromise folding. That uncertainty is real for this splice form, but every GOA row here is a gene-level IEA. The same rows are present on all horse ALDH7A1 entries. The gene-level claims are well grounded by PAINT node placement and human biochemistry. I therefore judged the rows as claims about the horse ALDH7A1 gene product: ACCEPT for AASA dehydrogenase, aldehyde dehydrogenase (NAD+), lysine catabolism and mitochondrion; KEEP_AS_NON_CORE for betaine-aldehyde DH, nucleus, Golgi membrane and plasma membrane. Cytosol is also KEEP_AS_NON_CORE, because the human cytosolic pool comes from the transcript lacking the transit peptide, which this product retains. Oxidoreductase activity is MODIFY to the two specific activities. The unverified activity of the exon-skipped product is stated in the description, the core functions, a suggested question and a suggested experiment.
Family review disagreement (interpro/panther/PTHR43521). The family review lists A0A9L0RRL6 as the SF1 representative and P49419 as the SF5 representative, with identical prose. Its summary describes the non-ALDH7 branch as "ALDH4/P5C-related". Human ALDH4A1 is in PTHR14516, not PTHR43521. The P5C branch inside PTHR43521 is ALDH12 (SF7, "DELTA-1-PYRROLINE-5-CARBOXYLATE DEHYDROGENASE 12A1"; SF6, L-glutamate gamma-semialdehyde DH), so the family review mislabels that branch. Using an exon-skipped gene model as the SF1 representative is also suboptimal; the mouse Q9DBF1 or bovine Q2KJC9 entries would be better representatives. I did not edit the family file.
Modules. modules/lysine_catabolism.yaml, modules/lysine_catabolism_saccharopine.yaml, modules/ferroptosis.yaml (all reference human ALDH7A1/P49419).