varKodes extend DNA barcoding to ultrafine-grained identity, place, and state

Peter J Flynn, Jackson Kehoe, T. Jonathan Davies, Lorène Gourguillon, Virginie Couturaud, Bruno de Medeiros, Charles Davis 

DNA barcoding has transformed biodiversity science, but locus-based markers often fail across taxa, at shallow phylogenetic depths, and in mixed or spatially structured samples. Here, we test whether varKoder, an alignment-free framework that converts low-coverage genomic reads into image-based genomic signatures (varKodes), can extend barcoding from taxonomic identification to broader inference of identity, place, and state. Across five public genomic and metagenomic datasets–domestic dogs, grape cultivars, soils, human skin microbiomes, and human stool microbiomes–varKodes resolved signals that span single-organism genomes and mixed microbial communities. They distinguished dog breeds and grape cultivars despite exceptionally recent divergence, hybrid ancestry, and complex pedigree histories. They also assigned skin microbiome samples to source individuals and body-site categories, demonstrating that host-associated microbial mosaics can function as individual-level genomic fingerprints. Beyond identity, varKodes recovered geographic provenance from soil metagenomes and distinguished skin microbiomes by country of origin, supporting metagenomic mosaics as location-specific signatures. Finally, varKodes classified inflammatory bowel disease status from stool metagenomes, providing a proof of concept for identifying biological state using the same kind of genomic signature. Performance was strongest for well-sampled or phylogenetically well-differentiated groups, indicating that denser reference data and retraining should improve accuracy. Together, these results demonstrate that DNA barcodes can be reconceived as retrainable, genome-wide signatures for supervised inference across a wide variety of biological scales, sample types, and applications.