Preprint proposes unified framework explaining why mixed-model GWAS behaves differently in livestock versus humans

A preprint posted to bioRxiv on 5 June 2026 presents a theoretical framework to explain a long-standing puzzle in livestock genomics: why mixed-model genome-wide association studies (GWAS) using the full genomic relationship matrix (full-GRM) yield only a handful of significant peaks even in datasets of hundreds of thousands of animals, whereas alternative approaches — including leave-one-chromosome-out (LOCO), numerator-relationship-matrix, and sparse-GRM methods — report many broader associations on the same data.

The authors argue that these contrasting behaviours can be traced to the low effective genomic dimensionality (Me) that characterises populations with small effective population size (Ne), such as commercial livestock breeds. In human GWAS, Ne is large enough that Me is correspondingly high; in cattle or pigs, Me is so low that the full-GRM absorbs signal that other methods leave accessible, fundamentally altering statistical power and fine-mapping resolution.

The framework provides an analytical basis for choosing between GWAS modelling approaches in low-Ne populations and has relevance to fine-mapping, heritability estimation, and the interpretation of polygenic architecture in livestock. The work has not yet been peer-reviewed.