The null distribution of stochastic search gene suggestion: a Bayesian approach to gene mapping

Bayesian methods continue to permeate genetic epidemiology investigations of genetic markers associated with or linked to causal genes for complex diseases. The attraction of these methods is an ability to capitalize on Bayesian priors to model additional complexity and information about the disease outside the specific data analyzed. It is well known that the larger the sample size, the more the Bayesian method with uninformative priors can be approximated by its Frequentist analogue. However, what is not known is how much impact the priors have on a Bayesian method when analyzing a null region of the chromosome. Here, we look at the impact of various prior values on stochastic search gene suggestion (SSGS) when analyzing a region of simulated chromosome 6 known to be unassociated with the simulated disease. SSGS is a recently developed Bayesian variable selection method tailored to investigate disease-gene association using case-parent triads. Our findings indicate that the prior probability values do affect false positives, and this study suggests values to calibrate the prior. Also, the sensitivity of the results to the prior probability values depends on two factors: the linkage disequilibrium between the marker loci examined, and whether this dependence is included in the model. In order to assess the null distribution we used the simulated data with the "answers" known.