Information Technology Reference
In-Depth Information
5
Discussion
The proposed method for detecting epistasis has the ability to determine which main
effects as well as which two-way interaction effects are present in a dataset as evidenced
by the simulation study. The method was applied to the
Arabidopsis thaliana
data and
no epistatic effects were found with respect to cotyledon opening angle. However, the
known locus for controlling cotyledon opening was detected, ATHCHIB2. The search
method was employed in a situation where the number of parameters in the full model
far exceeded the number of observations, but by placing restrictions on the parameter
space each model under consideration had sufficient degrees of freedom for estimating
parameters.
A study of epistatic models which do not require the first order terms to be present
should be considered as well. This may allow for better detection of epistatic effects as
the model search does not need to first add a main effect in order to later include the
epistatic term. In this case the model space would be reduced by
2
p
models. However,
if all other interaction terms are equally likely to be added to the model, the Metropolis-
Hastings step may have low acceptance probability and convergence of the
MC
3
algo-
rithm may be slow. In addition, any loci that have effects that are not in interaction with
other loci may not be detected. Hence, reducing the utility of the method.
Caution should be used when using restricted model spaces. The method works best
when it is believed that only a few loci control the trait of interest. In cases where it
is believed that a large number of loci control the trait of interest, especially when this
exceeds the restriction on the model space, then the search method maybe come very
ineffective at assessing both the main effect as well as the epstatic effects. Since the
models at the restriction boundary will have high posterior model probabilites it may
be difficult to move through regions of lower probability towards even more probable
models. In most cases in genetics it is believed that only a few loci control the trait of
interest.
References
1. Broman, K.W.: The Genomes of Recombinant Inbred Lines. Genetics 169, 1133-1146
(2005)
2. Broman, K.W., Speed, T.P.: A model selection approach for the identification of quantitative
trait loci in experimental crosses. J.R. Statist. Soc. B 64, 641-656 (2002)
3. Boone, E.L., Ye, K., Smith, E.P.: Assessment of two approximation methods for computing
posterior model probabilities. Computational Statistics & Data Analysis 48, 221-234 (2005)
4. Boone, E.L., Simmons, S.J., Ye, K., Stapleton, A.E.: Analyzing quantitative trait loci for the
Arabidopsis thaliana using Markov chain monte carlo model composition with restricted and
unrestricted model spaces. Statistical Methodology 3, 69-78 (2006)
5. Carlborg, O., Andersson, L., Kinghorn, B.: The Use of a Genetic Algorithm for Simultaneous
Mapping of Multiple Interacting Quantitative Trait Loci. Genetics 155, 2003-2010 (2000)
6. Cockerham, C.: An extension of the concept of partitioning hereditary variance for the anal-
ysis of covariances among relatives when epistasis is present. Genetics 39, 859-882 (1954)
7. Green, P.J.: Reversible jump Markov chain Monte Carlo computation and Bayesian model
determination. Biometrika 82, 711-732 (1995)
