Research

Our work is organized around one question: how to extract reliable genetic conclusions from very large samples of related individuals.

Theme 1

Mixed models at scale

Linear mixed models are the workhorse of quantitative genetics. We develop stochastic solvers that make mixed-model GWAS and genomic prediction practical for millions of genotyped animals, and REML estimation practical for hundreds of thousands.

The same machinery supports partitioning heritability across genomic annotations, which connects variance components to biology.

Software: SLEMM, MPH

Selected papers

Theme 2

Fine-mapping in related populations

Most fine-mapping methods assume unrelated samples, and their assumptions fail quietly in livestock data. We develop Bayesian fine-mapping methods for related samples, correct linkage-disequilibrium estimates for relatedness so that summary-statistic tools can be used, and apply these methods to sequence-imputed data on tens of thousands of animals.

Integrating functional genomic annotations helps prioritize variants and genes, and shows which kinds of genomic regions are enriched for genetic effects on complex traits.

Software: BFMAP, LD Adjuster, gemrich

Selected papers

Theme 3

Theoretical limits of genomic analysis

GWAS power, fine-mapping resolution, and genomic prediction reliability are usually studied separately. We show that all three are bounded by the effective dimensionality of the genome, which in turn depends on effective population size. This gives a common explanation for why some analyses succeed in humans and struggle in livestock, and the reverse.

We pair this theory with benchmarks that test how widely used methods behave under realistic genetic architectures.

Selected papers

Theme 4

AI-assisted quantitative genetics and breeding

We use AI where it adds something that classical models lack: genomic language models to predict variant effects across species, neural networks benchmarked against standard genomic prediction, and AI assistance for running mixed-model analyses.

DairyDelphi, an ongoing project, applies generative sequence models to the life histories of dairy cows recorded on farms, to forecast health, production, and fertility for breeding and management decisions.

Selected papers