Family-specific structure in infant gut viromes enables closed-set mother-child matching

Abstract The early-life gut virome differs so strongly between individuals that the variation is usually treated as noise. We show it is structured by family, specifically enough that a child's virus set points to their own mother. Across 2,664 catalogue metagenomes, 142 twin libraries and 284 shotgun metagenomes re-quantified from raw reads, we ranked 32 candidate mothers by viral overlap with each of 125 child samples from 32 families. Using all detected vOTUs, the true mother ranked first for 67 of 125 samples (53.6%; median rank 1), against a median sample-level accuracy of 2.4% after permuting maternal labels across families. The mean within-family accuracy was 52.0% (family bootstrap 95% confidence interval 37.2% to 66.6%), and the true mother ranked first at least once in 22 of 32 families. Specificity comes from where the background is empty, not where the signal is large: the own-mother excess is largest for viruses of intermediate prevalence (+ 0.64) and small for the rarest (+ 0.09), yet rare viruses discriminate best because unrelated pairs share almost none. The same asymmetry appears independently in the twin cohort, where rare-virus sharing separates co-twins from unrelated pairs at AUC 0.997. Matching is not merely taxonomic: median agreement across cohort-variable consensus positions in shared vOTUs was 1.000 for mother-child pairs against 0.768 for unrelated pairs, positive in all 24 evaluable families. These findings support family structure in infant viromes, with additional within-vOTU consensus-sequence concordance in the evaluable subset of shared viruses. The comparisons are closed-set and do not resolve transmission direction or separate maternal acquisition from continued shared exposure.