A framework for analyzing both linkage and association: an analysis of Genetic Analysis Workshop 16 simulated data

Abstract

We examine a Bayesian Markov-chain Monte Carlo framework for simultaneous segregation and linkage analysis in the simulated single-nucleotide polymorphism data provided for Genetic Analysis Workshop 16. We conducted linkage only, linkage and association, and association only tests under this framework. We also compared these results with variance-component linkage analysis and regression analyses. The results indicate that the method shows some promise, but finding genes that have very small (<0.1%) contributions to trait variance may require additional sources of information. All methods examined fared poorly for the smallest in the simulated "polygene" range (h2 of 0.0015 to 0.0002).

Figure 1

Associated SNPs account for linkage peaks. L-scores for MCMC oligogenic linkage analysis of LDL with (green) and without (red) simultaneous association testing at most causative SNPs. Including the causative SNPs reduces the linkage signal in most regions.

Figure 2

Agreement between methods within replicate. Comparison of linkage results for simulated LDL at Visit 1, adjusted for age and sex, for L-scores from oligogenic simultaneous segregation and linkage analysis (without association testing) and for LOD from variance-components methods. Vertical scales are not comparable, but note agreement of peak locations

Figure 3

Different signals in different replicates. LDL SOLAR linkage results. While several regions exhibited LOD > 1 in multiple replicates, no region did so in all replicates.

Figure 4

Linkage and Association vs. gene locations. HDL Visit 1 adjusted for age and sex analyzed SOLAR and regression analyses in 2-cM wide windows (moving in 1-cM increments). In Replicate 1, no evidence for linkage was found. In Replicate 2, a LOD score > 2 at ~80 cM was found.