Linear Regression GWAS

The simplest scan to do is a linear regression GWAS. The following model is first when doing a linear regression GWAS.

In the example data set, the phenotype, y, is coded 0,1. When GxEScanR sees the phenotype codes this way it assumes the outcome is binary and uses logistic regression. To perform a linear regression GWAS the binary option needs to be set to FALSE. The following shows how to do a linear GWAS along with the results.

The routine outputs the number of subjects used in the analysis and returns a data frame with the results.

lingwas1 <- gwas(data = covdata,
                 bdinfo = bdinfo,
                 binary = FALSE)
#> [1] "200 subjects have complete data"

The output can be redirected to output file that produces a plain test version of the results in a tab delimited file that can be read into R using the read.table routine. In this case, the gwas routine returns a value of 0.

outfile <- tempfile()
lingwas2 <- gwas(data = covdata,
                 bdinfo = bdinfo,
                 outfile = outfile,
                 binary = FALSE)
#> [1] "200 subjects have complete data"
lingwas2
#> [1] 0
lingwas2 <- read.table(outfile, header = TRUE, sep ='\t')

Linear Regression GWEIS

The gweis routine takes the same parameters as the gwas function but performs additional tests. The models fit for a linear regression GWAS are.

Note: When doing a GWEIS the interaction covariate is in the last column of the subject data frame.

In this test the minmaf option was used. When minmaf is specified the minor allele for a SNP must exceed minmaf to be test. Notice that only 5 SNPs are in the output data frame. This is because one of the SNPs has a minor allele frequency below 0.2.

lingweis1 <- gweis(data = covdata,
                   bdinfo = bdinfo,
                   minmaf = 0.2,
                   binary = FALSE)
#> [1] "200 subjects have complete data"

If the user is interested in see what happened to SNPs that weren’t included in the data frame, the skipfile option can be used. The skipfile value is the name of a file to write the skipped SNPs to. The skipfile option can be used along with the outfile option. The skip file is in the same format as the output file. Below is an example using the skipfile option.

skipfile = tempfile()
lingweis2 <- gweis(data = covdata,
                   bdinfo = bdinfo,
                   skipfile = skipfile,
                   minmaf = 0.2,
                   binary = FALSE)
#> [1] "200 subjects have complete data"

knitr::kable(skipsnps, caption = "Skipped SNPs")

The following table lists the reasons SNPs were skipped given the skipped value.

Logistic Regression GWAS

In this example, the phenotype is coded (0,1). The gwas and gweis routines check for this an will run logistic regressions if the outcome is coded (0,1) unless binary is set to FALSE. If the use wants to make sure the outcome is coded (0,1), the user may set binary to TRUE. In this case, if the outcome is not coded (0,1) an error is produced.

The following model is fit when doing a logistic regression GWAS.

loggwas1 <- gwas(data = covdata,
                 bdinfo = bdinfo,
                 blksize = 2,
                 binary = TRUE)
#> [1] "200 subjects have complete data"

In this example, the option blksize is used. When an analysis is run several SNPs are read in at one time. This saves disk time. The following are the default values for given the number of subjects. These values were chosen to keep the program running using less than 4GB of RAM. The user is allowed to specify a value up to twice the default value. Little performance gain is seen going with larger values. If the user enters 0 for blksize, the default value is used.

Logistic Regression GWEIS

A logistic regression GWEIS fits an additional 4 models that produce 7 more tests. 3 of these models use the the interaction covariate as the outcome. The following show all the models fit in a logistic regression GWEIS.

The following model is fit when doing a logistic regression GWAS.

Note: When doing a GWEIS the interaction covariate is in the last column of the data frame.

loggweis1 <- gweis(data = covdata,
                   bdinfo = bdinfo,
                   snps = 1:2,
                   binary = TRUE)
#> [1] "200 subjects have complete data"

covdata2 <- covdata
covdata2$e <- covdata2$e + 1
loggweis2 <- gweis(data = covdata2,
                   bdinfo = bdinfo,
                   snps = c("1:10001", "1:10002"))
#> [1] "200 subjects have complete data"