gls.approx.logistic: Calculate weights and transformed phenotype so that one...
In tobyjohnson/gtx: Genetics ToolboX

Description Usage Arguments Details Value Author(s) Examples

Logistic regression models are usually fitted by iteratively reweighted generalised least squares (GLS). This function formulates a GLS problem by calculating weights and rescaling the response variable such that a logistic regression analysis is approximated. This is equivalent to making a quadratic approximation to the likelihood and to performing a score test.

1
2
3

gls.approx.logistic(snpdata, leftvar, rightvars = NULL,
                    outvar = paste(leftvar, "star", sep = ""),
                    weightvar = "weight")

`snpdata`	a list with snpinfo and data, see snpdata.
`leftvar`	the name (in `snpdata$data`) of the response variable for the logistic regression. Must be 0/1 or NA for all individuals.
`rightvars`	the names (in `snpdata$data`) of the explanatory variables for the logistic regression.
`outvar`	the name (to be added to `snpdata$data`) of the transformed phenotype that is the response variable in the GLS problem.
`weightvar`	the name (to be added to `snpdata$data`) of the weight variable in the GLS problem.

An intercept term is always included and there is currently no way to override this.

The specified model is printed as a check. MLE parameter values are estimated under the specified model by calling glm. These are used to compute weights and a transformed phenotype such that one iteration of generalised least squares constitutes a score test for additional terms that might subsequently be added to the model.

This function works on the (usually phenotypic) columns in the $data slot of snpdata.

Returns snpdata with additional columns for the weights and transformed phenotype.

The formula used to find an MLE is printed.

Toby Johnson Toby.x.Johnson@gsk.com

data(mthfrex)
mthfrex <- gls.approx.logistic(mthfrex, "HTN", c("SexC", "Age"))
xtwx <- make.moments2(mthfr.params, c("HTNstar", "SexC", "Age"), mthfrex,
                      weightvar = "weight")
est.moments2(xtwx, "HTNstar", c("ONE", "rs6668659_T", "rs4846049_T",
                                "rs1801133_G", "SexC", "Age"), vscale=1)