APSampler is a tool that allows multi-locus and multi-level association analysis of genotypic and phenotypic data. The goal is to find the allelic sets (patterns) that are associated with phenotype. The main difficulty of such a task is, given the multiple loci and multiple alleles, the number of all possible classifiers tends to be extremely large. Therefore, Monte Carlo Markov Chain method is applied to reduce the space of solutions and sample only from regions where it is likely to find a good classifier.
Once a set of classifiers is found, there is a problem to validate the results, and this is done using a number of well known methods. In case of single disease level, the resulting classifier divides the space of healthy and ill individuals, and the result is represented in a classic Fisher table. Odds ratio and Fisher's p-value are calculated if applicable. Also, Kruskal's gamma and the corresponding p-value can be calculated. After each pattern in the output is described by a p-values set of different multiple-hypothesis corrections, including permutation tests. The permutation procedure is highly paralleliseable, the examples of parallel scripts are in the package.
We use a pair of statistical tests for reporting interaction study results. These two tests are similar to the OR/CI and the exact Fisher test for association and implemented those in a software program. The pair is the SF test described by (Cortina-Borja et al. 2009) and the Fisher-like interaction numeric test (FLINT), which by all means is similar to the 3-way interaction exact test, introduced by (White, Pesner, and Reitz 1983) for use in sociology. More information cam be found on Readme Episasis wiki page and on Flinte (Fisher-Like Intraction Numeric Test software hom page
The authors would like to thank NIH/NLM (LM008932), UEPHA*MS FP7 Marie Curie Initial Training Network (FP7/2007-2013, grant agreement 212877), Russian Foundation for Basic Research (11-04-02016-a and 13-04-40279-H-KOMFI) for support of this work.
APSampler is an open-source project, it's source code is available for free on sourceforge. Documentation in wiki-style is available via this link.
Application citations are available here
Feel free to contact us.