Datensatz

Zusammenfassung

Background / Purpose:
Genetic association studies have become an integral tool to understand how genetic variation effects a growing selection of phenotypes. These may range from molecular phenotypes, such as gene expression, to global phenotypes such as growth rate or floweringtime of plants. Thanks to recent advances in sequencing technology, large-scale genotype information for hundreds, soon thousands of humans, plants and animals are now available or soon to be released. If new phenotypes are being studied, it is rarely feasible to phenotype all the individuals, due to cost and or time constraints. To solve this emerging problem of appropriate sample selection we propose and investigate experimental design by a principled information criterion.
Main conclusion:
In a retrospective real-data study on a flowering-time QTL study of 166 individuals of A. thaliana (Atwell et al 2010), we demonstrate the viability of experimental design in genome-wide association studies, achieving a significant improvement over a non-optimised study layout and simple baseline selection criteria. (See figure on poster: “RMSE on independent test set”: Experimental design curves for the Bayesian Lasso and various selection criteria compared to random selection on the A. thaliana flowering-time dataset. ALC denotes the area under the learning curve.)
Disclosures:
No relevant conflicts of interest declared.