Bacterial and archaeal COI homologues from GTDB R232: a reference set for off-target screening

This dataset provides nucleotide sequences of bacterial and archaeal homologues of mitochondrial cytochrome c oxidase subunit I (COI/COX1), derived from GTDB R232 species representative genomes. It supports the detection of non-target prokaryotic sequences in COI metabarcoding data and the screening of COI reference databases for records of possible prokaryotic origin. Nomenclature and scope: "COI homologues" refers here to subunit I of the heme-copper oxidase superfamily: cytochrome c oxidases, quinol oxidases and the related nitric oxide reductases, which the build does not separate from the oxidases. The corresponding bacterial and archaeal genes have different names, including coxA, ctaD, cyoB, qoxB and ccoN. "COX1" in the release identifier, the file names and the README is shorthand for this broader homologue set, rather than a universal gene symbol or an assertion that every sequence encodes the same enzyme. The genes are GTDB's own Prodigal predictions, found with the Pfam model PF00115 and kept when they score at least 100 bits against it and at most one of the six ligand histidines is replaced. Each gene was then checked for its reading frame and its GC content against its genome, and compared with the MIDORI2 records that are certainly mitochondrial. The MIDORI2 GB273 records at least 90 % identical to a gene that passed these checks were examined next: those for which the gene's relatives or the record's tryptophan codons pointed to a bacterial origin, and neither to a mitochondrial one, were marked for removal as bacterial COI homologues filed under eukaryote names, and the others were listed. Last, each gene was searched against the cleaned MIDORI2 and against the genes of the other genomes; genes for which the search found a MIDORI2 record at least 80 % identical and closer than any gene of another genome were left out or removed, as probable eukaryotic sequences binned into bacterial genomes. This included the matching genes of 17 of the examined records. Contents: 242,698 whole genes and 31,710 fragments of at least 200 amino acids from 123,356 genomes; a table of every candidate gene with the outcome of every check; the list of the 186 MIDORI2 GB273 records examined in this way, 170 of them marked for removal; the build scripts (scripts.zip), the conda environment and the HMM checksums to rebuild it. The README gives the counts per step, the sensitivity of the counts to twelve of the thresholds, and the results of a final test on data not used to set the rules. Built for MetaFlux (COI marker); the method is described step by step in the MetaFlux manual, page "Bacterial COI homologues (reference set)": https://ilivius.github.io/MetaFlux/amplicon/markers/bacterial-cox1/ Zenodo keeps no folders: the build scripts are in scripts.zip. Unzip it in the download folder before checking all files with sha256sum -c SHA256SUMS.