Evolutionary replay reveals lineage-associated predictability of duplicate-gene retention across independent whole-genome duplications

Whole-genome duplication repeatedly exposes ancestral gene lineages to duplicate retention and loss. Although preferential retention is established, whether lineage identity carries transferable information, and which biological properties account for that information, remain separate questions. We compared retention ranks across independent duplications using orthology coordinates constructed independently of target outcomes. Three plant events showed replay (Treplay = 0.210; bootstrap 95% confidence interval, 0.172 to 0.248; permutation P = 1/100,001). A frozen plant score prospectively predicted Apple/Pear retention ({rho} = 0.169; n = 373; 95% confidence interval, 0.069 to 0.262; P = 0.000470). Transfer to yeast was supported, whereas teleost transfer remained unresolved. Independent teleost-Stylommatophora replay was positive but below its prespecified strong-effect threshold; a strict plant-animal scalar comparison remained unresolved. We then prospectively defined tests of dosage-related response, protein-sequence constraint, expression magnitude and expression breadth, followed by a frozen multivariate analysis. Dosage-related proxies, sequence constraint and expression breadth were unsupported under their tested definitions. Conserved expression magnitude was associated with retention across the reference plant events, but its attenuation of replay was unresolved. The multivariate model did not establish reliable prediction of held-out events superior to all single-property models. Thus, gene-lineage identity carries partially transferable retention information, but the tested conserved properties do not demonstrably account for it. Predicting retention and explaining replay are distinct biological problems.