Efficient encoding of large antigenic spaces by epitope prioritization with Dolphyn

The human gut hosts trillions of bacteriophages — viruses that infect bacteria — whose antigens make up an enormous "antigenic space." Profiling antibody responses across that space with phage immunoprecipitation sequencing (PhIP-Seq) is in principle straightforward, but in practice the peptide library needed to tile every gut-phage protein is so large that the assay becomes too expensive to deploy at cohort scale.

Dolphyn is the authors' answer to this scaling problem. It does two things at once. First, a random-forest classifier scores every 15 amino acid window of every input protein on its likelihood of containing a linear B-cell epitope, based on amino acid composition; only the highest-probability windows are kept. Second, an "epitope-stitching" step concatenates three high-probability 15-mers from one protein into a single synthetic 45 amino acid peptide, joining them with flexible Gly–Gly–Gly–Gly–Ser linkers so that the immunological behavior of each fragment is preserved. The compressed library still represents three times as many epitope candidates per oligonucleotide synthesized.

The team validated the design on the human virome (using VirScan reactivity data for EBV, rhinovirus, and CMV peptides) and on enterovirus reactivities from the DAISY pediatric cohort, and showed that Dolphyn captured the same immune signal as a much larger uniformly-tiled library while consuming roughly three-fold fewer peptides. Then they applied Dolphyn to a database of 142,810 gut bacteriophage genomes, built a 106,762-peptide library, and used it to characterize the gut-phage antibody response in a cohort of healthy adults. The headline finding: people make antibodies against their gut phages, and the dominant signal comes from E. coli-infecting Myoviridae — placing gut phages firmly inside the gut–immune axis as a recognized antigenic compartment.