Synteny over similarity: human–AI curation of the caspase family in allopolyploid carp genomes

Allopolyploid carps (Cyprinus carpio, Carassius auratus, Carassius gibelio) are the subject of a substantial functional literature, and are now supported by high-quality reference genomes and annotations. Drawing on those resources requires navigating the consequences of a hybridization event 10–13 million years ago that left these species with two subgenomes and homeologous A/B gene copies. Annotation names alone do not distinguish homeologs from paralogs. Annotation files record expert judgement, but resist interpretation when many loci must be held in view. We describe a synteny-based approach in which a human curator and a capable AI agent work jointly on public annotation (GFF3) and protein sequence (FASTA) files, with the agent performing mechanical operations guided by researcher judgement. Applied to the caspase family, 33 annotated Prussian carp caspases resolved into 16 syntenic slots across 12 homeologous chromosome pairs; at five slots a homeolog was absent from one subgenome. The same core architecture was recovered in goldfish and common carp, with lineage- and annotation-specific differences at individual slots. The result is a subgenome-resolved description of the caspase family, in which relationships are established from locus position rather than from annotation labels, and asymmetries between subgenomes are located at specific coordinates.