R: Summarize variable selection results in a single plot.

plot.varbvs {varbvs}

R Documentation

Summarize variable selection results in a single plot.

Description

Generate a single plot that summarizes the results of fitting the Bayesian variable selection model to the data. When the variables are genetic markers, the groups are chromosomes, and the posterior probabilities are plotted on the vertical axis (typically on the logarithmic scale), the figure resembles a "Manhattan plot" typically used to summarize the results of a genome-wide association study or quantitative trait locus (QTL) mapping study.

Usage

  ## S3 method for class 'varbvs'
plot(x, score, groups, vars = NULL, var.labels,
  draw.threshold = NA, gap = 0,col = "midnightblue", pch = 20,
  scales = NULL, xlab = "variable", ylab = "posterior probability",
  main = "fitted varbvs model: variable selection results",
  abline.args = list(lty = "dotted",col = "orangered"),
  vars.xyplot.args = list(pch = 20,col = "magenta"),
  vars.ltext.args = list(col = "black",pos = 4,cex = 0.5), 
  par.settings = list(par.main.text = list(font = 1,cex = 0.8),
                      layout.heights = list(axis.top = 0, axis.bottom = 0)),
  ...)

Arguments

`x`	Output of function `varbvs`.
`score`	Value to plot on vertical axis. Must be a numeric vector with one entry for each variable. If missing, the posterior inclusion probability for each variable is plotted in the vertical axis, in which this probability is averaged over hyperparameter settings, treating `x$logw` as (unnormalized) log-marginal probabilities. As alternative, set `score` to the posterior inclusion probabilities that ignore correlations, using `varbvsindep`.
`groups`	Group the variables in the plot according to this argument. This must be a vector with one entry for each variable. If missing, all variables are treated as a single group. This is useful for grouping the genetic markers by chromosome in a genome-wide association study.
`vars`	Indices (type `integer`) or names (type `character`) of variables to highlight and label. By default, `vars = NULL`, meaning no variables are highlighted.
`var.labels`	Labels to accompany the highlighted variables only. If missing, labels are retrieved from `x`. If `var.labels = NULL`, no labels are plotted.
`draw.threshold`	Plot a horizontal line at this location on the vertical axis.
`gap`	Amount of space to leave between each group of variables in the plot.
`col`	Argument passed to `xyplot` specifying color of points.
`pch`	Argument passed to `xyplot` specifying symbol type.
`scales`	Argument passed to `xyplot` specifying how x- and y-axes are drawn.
`xlab`	Argument passed to `xyplot` specifying horizontal axis title.
`ylab`	Argument passed to `xyplot` specifying vertical axis title.
`main`	Argument passed to `xyplot` specifying main plot title.
`abline.args`	Additional arguments passed to `panel.abline` specifying how to draw the horizontal line at the location specified by `draw.threshold`.
`vars.xyplot.args`	Additional arguments passed to `xyplot` for drawing the highlighted variables.
`vars.ltext.args`	Additional arguments passed to `ltext` for specifying how the labels of the highlighted variables are drawn in the plot.
`par.settings`	Argument passed to `xyplot` specifying additional adjustments to the plot settings.
`...`	Additional arguments passed to `xyplot` for drawing the un-highlighted variables.

Details

Note that plot.varbvs uses function xyplot from the lattice package, and as.layer from the latticeExtra package.

Value

An object of class "trellis" generated by functions xyplot and as.layer.

Author(s)

Peter Carbonetto peter.carbonetto@gmail.com

References

P. Carbonetto and M. Stephens (2012). Scalable variational inference for Bayesian variable selection in regression, and its accuracy in genetic association studies. Bayesian Analysis 7, 73–108.