freqMAP (version 0.2)

freqMAP-package: Frequency Moving Average Plots (MAP) of Multinomial Data by a Continuous Covariate

Description

This package can be used to estimate a frequency moving average plots (MAP) from multinomial data and a continuous covariate. The frequency MAP is a moving average estimate of category frequencies, where frequency means and posterior bounds are estimated. Comparisons of two frequency MAPs as well as odds ratios can be plotted.

Details

Package:

freqMAP

Type:

Package

Version:

0.1

Date:

2008-09-24

License:

GPL (>= 2)

This package allows one to visualize frequency data (e.g. genotype) as a function of a continuous covariate (e.g. age). Two populations can be visualized on the same plot and inference can be made about significant differences in the frequencies and odds ratios comparing two unmatched populations. This can be useful for exploring possible confounding effects in a case control study. In this application, one frequency MAP would be used for cases and another MAP for controls. Then one would plot the frequency MAP of, say, genotype frequencies as a function of age in both groups. Interactions between age and genotype would show up as non-parallel frequency MAPs in the two groups. This information can then be used for formulating interactions in the subsequent logistic regression modeling.

At each of a pre-specified set of values of the continuous covariate, the frequency data is modeled as Multinomial with an unknown true category frequency vector. The prior on the true category frequency vector is assumed to be uniform ( Dirichlet(1,...,1) ). Then samples are generated from the Dirichlet posterior distribution on the true category frequency vector. The central 95% posterior interval on all true category frequencies are estimated from the posterior samples.

The function freqMAP creates a freqMAP object and samples from the posterior on true category frequencies. The object can then be plotted using plot.freqMAP. The posterior distributions from two unmatched populations contained in two freqMAP objects can be compared using posterior.comparison.freqMAP. plot.freqMAP can also plot two populations at once as well as odds ratios comparing the two populations.

The example below analyzes a typical dataset one would find in a genetic association study, where genotypes are collected for a group of probands and controls of varying ages. See genotypebyage for more details on the dataset. The example code shows how to generate MAP plots by genotype and by allele.

References

Payami, H., Kay, D.M., Zabetian, C.P., Schellenberg, G.D., Factor, S.A., and McCulloch, C.C. (2009) "Visualizing Disease Associations: Graphic Analysis of Frequency Distributions as a Function of Age Using Moving Average Plots (MAP) with Application to Alzheimer's and Parkinson's Disease", Genetic Epidemiology

Examples

data("genotypebyage")

####An analysis by genotype

#Create a freqMAP object for probands and controls, respectively.
fm.cont <- freqMAP(genotypebyage[genotypebyage$subject.type=="control",c("age","genotype")],
                   x=seq(25,95,by=5),x.label="age",hw=5)
fm.prob <- freqMAP(genotypebyage[genotypebyage$subject.type=="proband",c("age","genotype")],
                   x=seq(25,95,by=5),x.label="age",hw=5)

#Plot the two MAPs overlaid. 
plot(fm.cont,fm.prob,legend=c("controls","probands"),show.p.value.legend=TRUE)


####An analysis by allele

#Reformat the dataset to stack the two alleles into one column. The
#resulting dataset has twice as many rows as genotypebyage
allelebyage <- rbind(cbind(genotypebyage[,c("age","subject.type")],
                           allele=genotypebyage$allele1,stringsAsFactors=FALSE),
                     cbind(genotypebyage[,c("age","subject.type")],
                           allele=genotypebyage$allele2,stringsAsFactors=FALSE))

#Create a freqMAP object for probands and controls, respectively.
fm.cont <- freqMAP(allelebyage[allelebyage$subject.type=="control",c("age","allele")],
                   x=seq(25,95,by=5),x.label="age",hw=5)
fm.prob <- freqMAP(allelebyage[allelebyage$subject.type=="proband",c("age","allele")],
                   x=seq(25,95,by=5),x.label="age",hw=5)

plot(fm.cont,fm.prob,legend=c("controls","probands"),show.p.value.legend=TRUE)