Bayesian Association of Multiple
SNP Effects (BAMSE)
BAMSE is a software for performing association studies of
multiple SNPs and environmental factors. We have developed a method based on
Bayesian statistics that can model interactions for a large number of SNPs and
environmental risk factors while accounting for the multiple testing problem.
More specifically we have developed a Markov Chain Monte Carlo method that
allows for identification of sets of SNPs and environmental factors that when
combined increase disease risk or change the distribution of a quantitative trait.
In this method, combinations of genetic and environmental genetic factors
define risk sets. Individuals with genotypes that are members of a risk set
have modified distributions of disease risk or quantitative trait value.
Phenotypic traits are modelled using normal distributions (quantitative traits)
or binary traits (case control). A Markov chain is established with state space
on the set of all possible risk sets and parameters (e.g. mean and variance) of
trait value distributions for each risk set. The stationary distribution of the
Markov chain is given by the posterior density of risk sets and trait value
distributions. Statistical inferences are based on this posterior distribution.
The Markov chain is simulated by the Metropolis-Hastings algorithm using reversible
jumps, to jump between risk sets. This method differs fundamentally from
previous approaches by entertaining non-linear models and by addressing the
multiple testing problem in a computationally and statistically efficient
manner. The method is developed for unrelated individuals.
Any comments, questions or bugs please contact me at albrecht@binf.ku.dk
Changes
Also allows binary traits (case control)
New handling of missing data