Generative Artificial Intelligence in Population Genetics

Understanding the demographic history of populations and species is essential, as it plays a critical role in conservation and breeding programmes for both wild and domesticated species. However, inferring demographic history from genetic data faces the problem of model non-identifiability, where different demographic histories fit the genetic data equally well. This issue is amplified by the use of summary statistics instead of complete genetic data. Addressing the model non-identifiability problem is fundamental for improving the precision of demographic inference, particularly in complex cases involving multiple independent admixture events or barriers to gene flow.This project aims to resolve the problem of model non-identifiability in demographic inference by developing a generative artificial intelligence (AI) tool that can uncover histories that are indistinguishable when a particular summary statistic is used. This generative AI will serve three core purposes: (1) identify non-identifiable demographic histories, helping researchers better understand the range of possible historical events that could explain the data; (2) apply this tool to clarify complex demographic histories in domesticated species, such as cattle, providing more accurate insights into their genetic background and informing breeding programmes to maintain genetic diversity; and (3) detect barriers to gene flow in recently diverged species, offering deeper insights into local adaptation and speciation processes valuable for conservation.By addressing the model non-identifiability problem and applying this approach to key biological questions, this project has the potential to transform the study of demographic history in population genetics and open new perspectives for research on species evolution and adaptation.