GP4PG: A metaheuristic approach to demographic inference using genetic programming

The use of genomic data to make evolutionary inferences has revolutionized our understanding of the processes that shape genetic diversity and adaptation across species. By analyzing genetic sequences, we can trace the evolutionary history of populations, species, and entire lineages, providing deeper insight into how evolutionary forces have influenced the genetic makeup of populations over time. Traditional demographic inference methods rely on summary statistics, which can either be selected by researchers or generated using deep learning techniques. While these deep learning approaches are useful for defining optimal summary statistics for demographic models, their black box nature can make them difficult to interpret. Without a clear understanding of how these algorithms function, making accurate evolutionary inferences becomes challenging. Traditional demographic modelling describes each tested model as a combination of topology and demographic events. However, the ascertainment of such models is often subject to bias, primarily due to the subjective selection of demographic events by researchers, particularly when population substructure is involved. To address these limitations, we developed the Genetic Programming for Population Genetics (GP4PG) methodology, an innovative approach to modelling complex demographic scenarios by applying genetic programming. GP4PG automates the exploration of parameter-free demographic models by coding demographic relationships as graphical structures, enabling optimization through mutation and selection. GP4PG minimizes model bias by only relying on the topology of demographic models rather than predefined parameter assumptions

Mar 12, 2025 12:00 AM