¹Department of Mathematics and Statistics, Rolf Nevanlinna Institute, University of Helsinki, Helsinki, Finland

Correspondence: Dr MJ Sillanpää, Department of Mathematics and Statistics, Rolf Nevanlinna Institute, University of Helsinki, P.O. Box 68, FIN-00014 Helsinki, Finland. E-mail: mjs@rolf.helsinki.fi

Received 8 September 2008; Revised 9 April 2009; Accepted 14 April 2009; Published online 20 May 2009.

Abstract

For small pedigrees, the issue of correcting for known or estimated relatedness structure in population-based Bayesian multilocus association analysis is considered. Two such relatedness corrections: [1] a random term arising from the infinite polygenic model and [2] a fixed covariate following the class D model of Bonney, are compared with the case of no correction using both simulated and real marker and gene-expression data from lymphoblastoid cell lines from four CEPH families. This comparison is performed with clinical quantitative trait locus (cQTL) models—multilocus association models where marker data and expression levels of gene transcripts as well as possible genotype expression interaction terms are jointly used to explain quantitative trait variation. We found out that regardless of having a correction term in the model, the cQTL-models fit a few extra small-effect components (similar to finite polygenic models) which itself serves as a relatedness correction. For small data and small heritability one may use the covariate model, which clearly outperforms the infinite polygenic model in small data examples.