The effective number and profile of causative markers in GWAS

Over the past decade, animal databases have amassed abundant genomic data, with gradual progression from us- ing thousands to using millions of markers. Against expectations, this increase in marker density only marginally increased the predictivity of the standard quantitative genetic models. To reason for this, we used stochastic sim- ulations and statistical modelling to understand the marker profile around the quantitative trait nucleotides (QTN) in populations with a small effective population size (Ne). We applied the single-step genome-wide association to simulated datasets with varying Ne and number of genotyped individuals with phenotypes. We decomposed resulting Manhattan plots into signals from the QTN itself, QTN profile, relationships, and noise. In this sense, the QTN profiles can be understood as pairwise linkage disequilibrium curves, with width inversely related to Ne. Thus, in larger populations, QTN profiles were narrow, which, coupled with weaker relationships, facilitated QTN identification (with ample phenotypic data). Conversely, smaller populations yielded wider QTN profiles that blended with relationship signals and noise, resulting in reduced GWAS resolution and QTN identification. In terms of predictivity, predictions in large populations required dense marker panels and QTN or nearby markers identified. In contrast, in smaller populations, medium marker density sufficed as the QTN profiles captured the QTN without its explicit identification.