Predictive Evolutionary Genomics: Principles, Validation, and Practice

Daniel Ortiz-Barrientos, Maddie Ellen James, Liu Yang, Dan Gabriel Bock, Moises Exposito-Alonso, Loren Henry Rieseberg

Contemporary evolution occurs on observable timescales, enabling prospective evolutionary forecasting with quantitative frameworks rather than only retrospective inference. We propose a unified probabilistic framework that integrates three approaches linked to detectability windows through time. Trait-based models use multivariate quantitative-genetic equations to project correlated phenotypic responses over ~5–20 generations while the G-matrix is locally stable. Allele-based analyses model frequency dynamics at identifiable loci over ~20–100 generations, when selection outpaces drift and sampling error. For longer horizons, composite adaptation scores aggregate many small effects to support 100+-generation projections under novel environments. These windows reflect detectability and parameter stability, not genetic architecture. Bayesian inference integrates genomic, phenotypic, and environmental evidence to yield probabilistic predictions with explicit uncertainty. Experimental evolution, historical herbarium series, and reciprocal transplants provide independent validation of forecast skill and transferability. Validated forecasts can guide conservation programs (genomic vulnerability and assisted gene flow), breeding programs (multi-generational, climate-aware selection), and ecosystem management (species-composition planning). By quantifying and propagating uncertainty, this programme shifts evolutionary biology from largely descriptive synthesis toward predictive practice.