Introduction to CASMAP

Dean Bodenham

27 June 2018

Introduction

The CASMAP package provides methods for searching for combinatorial associations inbinary data while taking categorical covariates into account. There are two main modes: the methods either search for region-based mappings or for higher order epistatic interactions.

Creating CASMAP objects

To create a CASMAP object, it is necessary to specify the mode. The first example below creates an object that will perform a region-based GWAS search, and then sets the target family-wise error rate to 0.01.

The next example shows how to create an object that will search for arbitrary combinations, i.e. a higher order epistatic search. Note that it is also possible to set the target family-wise error rate when constructing the object by setting alpha.

By printing the object, one can see certain information. The field Maximum combination size = 0 indicates that combinations of all possible length will be considered. In future versions, it will be possible to limit this number, for example to combinations of maxmimum length 4.

## CASMAP object with: 
##  * Mode = higherOrderEpistasis 
##  * Target FWER = 0.01 
##  * Maximum combination size = 0 
##  * No input files read

Reading in the data files

Once the object is created, the next step is to read in the data files. The readLines command is used, and paths to the data files should be specified for the parameters genotype_file, phenotype_file and (optionally) covariate_file. We have provided example data files with the package, as well as functions to easily get the paths to these data files:

## CASMAP object with: 
##  * Mode = regionGWAS 
##  * Target FWER = 0.05 
##  * Maximum combination size = 0 
##  * Input files read 
##  * Covariate = TRUE

Data format

Note that the CASMAP methods expect the data file to be a text file consisting of space-separated 0s and 1s, in an \(p \times n\) matrix, where each of the \(p\) rows is a feature, and each of the \(n\) columns is a sample/subject. The labels and covariates files are single columns of \(n\) entries, where each entry is 0 or 1. To see an example of the data format, take a look at the included example files, the paths to which are given by the commands getExampleDataFilename, getExampleLabelsFilename and getExampleCovariatesFilename:

In future versions the PLINK data format will be supported.

Executing the algorithm

Once you have read in the data, label and covariates files, you are ready to execute the algorithm. Simply use the execute command. Note that, depending on the size of your data set, this could take some time.

Extracting the results

There are two main sets of results:

  1. Summary results
  2. Information on significant regions/significant interactions

The summary results provide information on how many regions/interactions were processed, how many are testable, and what are the significance and testable thresholds:

## $n.int.processed
## [1] 18193
## 
## $n.int.testable
## [1] 16426
## 
## $testability.threshold
## [1] 2.630268e-06
## 
## $target.fwer
## [1] 0.05
## 
## $corrected.significance.threshold
## [1] 3.043955e-06

It is also possible to write this information to file directly using the writeSummary command.

The significant regions lists all the regions that are considered significant. However, it is possible that these regions overlap into clusters. The most significant regions in these clusters can be extracted using the getSignificantClusterRepresentatives command. In the example below, there is only one significant regions, so it is its own cluster representative:

##   start end    score odds_ratio       pvalue
## 1    99 102 24.56281   16.12821 7.192676e-07
##   start end    score odds_ratio       pvalue
## 1    99 102 24.56281   16.12821 7.192676e-07

Note that the \(p\)-value and odds ratio for the regions/representatives is provided along with the location.

For the higherOrderEpistasis mode, the method getSignificantInteractions should be used (and there are no cluster representatives).

Other examples

It is also possible to perform a search without any covariates:

## CASMAP object with: 
##  * Mode = regionGWAS 
##  * Target FWER = 0.05 
##  * Maximum combination size = 0 
##  * Input files read 
##  * Covariate = FALSE

Setting the encoding method

The binary data could be encoded with either a dominant or recessive encoding. The default for readLines is dominant, but it is also possible to specify the coding explicitly:

Future releases

Note that future versions of the package will include the option to read PLINK files, and the option to set the maximum combination length.