Deconvoluting virome-wide antibody epitope reactivity profiles

VirScan is a programmable phage-display assay that displays roughly 100,000 short peptides spanning the entire human virome and then measures which of those peptides are bound by antibodies in a patient’s serum. The promise is breathtaking — one assay, one drop of blood, the cumulative record of viral exposure. The problem, until this paper, was that many of those peptides come from highly conserved regions shared across related viruses. A patient infected with herpes simplex virus 1 (HHV-1) might also score positive for cytomegalovirus (HHV-5) simply because their antibodies recognise a peptide motif both viruses carry. Earlier VirScan analytical pipelines counted hits without correcting for this cross-reactivity, and they did not adjust for the fact that some viruses are represented by far more peptides in the library than others.

AVARDA — the AntiViral Antibody Response Deconvolution Algorithm — addresses both gaps. It first builds a sequence-alignment graph that records every peptide-to-peptide and peptide-to-virus relationship in the library. It then assigns each observed binding event probabilistically across the candidate viruses that could explain it, and tests whether the cumulative evidence for any given virus exceeds what would be expected by chance. The output is a per-virus probability score and an effective-breadth measurement that resists the classic “one shared motif inflates ten species” failure mode — the same conceptual move that gave shotgun metagenomics species-level resolution a decade earlier.

The authors validated the approach on three cohorts. In a set of encephalitis patients with paired acute and convalescent sera, AVARDA correctly identified the responsible herpesvirus (HHV-1, HHV-3, HHV-4, HHV-5, or HHV-6) in cases that prior testing had either confirmed clinically or left undiagnosed — yielding a 44% improvement in viral encephalitis diagnosis. Applied to a type 1 diabetes cohort, AVARDA implicated enterovirus exposure as a candidate environmental trigger consistent with longstanding hypotheses in the field. Applied to systemic lupus erythematosus, it surfaced herpesvirus signals consistent with the EBV-autoimmunity literature. The implication for translational researchers studying immune-mediated disease is direct: groups running virome serology on autoimmune cohorts should expect to revisit historical hit lists once probabilistic deconvolution is applied, and acute-versus-convalescent paired serum is the high-value design for unlocking the analytical headroom this method provides.