Population stratification and the limits of polygenic inference
How much population structure survives standard correction in large biobanks, and what the residual does to polygenic scores and tests for selection.
Research
My research runs across two connected programs organized around a shared question: how much does a population’s history determine what it can do next — and can we know that before it matters?
Program I — Evolutionary Genetics of Adaptation and Persistence
For readers from evolutionary biology, ecology, integrative biology, and conservation genetics.
This program studies what constrains evolutionary response in wild and historical populations: whether a population can adapt and decline simultaneously; how much of its heritable variation selection can actually use; and what a population crash costs in the direction of remaining variation. It draws on long-term pedigree records from monitored vertebrate populations and preindustrial human genealogies, forward-time simulation across finite genomes, and analytical quantitative genetics. Most directly relevant threads: Mutational drought and the fate of small populations, Evolutionary predictability in human populations, Linked selection and the neutral diversity we actually observe.
Program II — Population History and Human Genetic Architecture
For readers from genetics, human genetics, genomics, and molecular and cell biology.
This program asks what population history did to the variation underlying complex traits that medicine now predicts from: how much sits in rare variants selection held at low frequency; whether ancestry stratification that survives PCA correction propagates coherently into polygenic scores and selection tests; and whether a trait’s genetic basis turns over between populations fast enough that a score cannot follow it. It draws on random matrix theory, allele frequency spectra, and biobank-scale GWAS data. Most directly relevant thread: Population stratification and the limits of polygenic inference. The quantitative-genetic and linked-selection work in Program I provides the analytical foundation that connects both programs — population history acting on variation twice, in wild populations and in clinical prediction alike.
Mutational drought and the fate of small populations
Small populations are usually thought to decline because deleterious mutations accumulate. The shortage of beneficial mutations turns out to matter more.
Linked selection and the neutral diversity we actually observe
Published heritable fitness variances are too small to explain the diversity wild populations have lost, and too large to be maintained at equilibrium. No rescaling of the estimates resolves both at once.
Evolutionary predictability in human populations
Deep genealogical records let us watch selection act on human life-history traits over centuries. What they show is that heritability plus selection does not predict the outcome.