OCplus (version 1.46.0)

average.fdr: Average a two-dimensional local false discovery rate

Description

This function averages two-dimensional local false discovery rates as computed by fdr2d for binned values of the first test statistic and across the values of the second test statistic. The result can easily be plotted and should be comparable to the one-dimensional fdr as provided by fdr1d, provided that the smoothing parameters were chosen suitably.

Usage

average.fdr(x, breaks)

Arguments

x

an object returned by fdr2d.

breaks

breaks defining intervals into which the first test statistic is binned; by default the same values that were used by fdr2d.

Value

A matrix with two columns tstat and fdr.local.

Details

Assuming that we have smoothed the estimate suitably and have chosen the proportion of non-dffierentially expressed genes suitably, we should get very much the same results from fdr2d as from fdr1d when we average across the logarithmized standard errors, see Examples.

The averaging is done across the estimated values for the actual genes; this corresponds to a weighted mean of the smoothed estimates on a grid, where the weight is proportional to cell frequencies.

Note that it is usuually easier to get a good match in the tails of the curves than in the center, which is okay, as this is where we want to estimate the fdr reliably.

References

Ploner A, Calza S, Gusnanto A, Pawitan Y (2005) Multidimensional local false discovery rate for micorarray studies. Submitted Manuscript.

See Also

fdr2d, fdr1d

Examples

# Create res1d
example(fdr1d)

# Compute fdr2d using the p0
res2d = fdr2d(xdat, grp, p0=p0(res1d))

# Show it
par(mfrow=c(2,1))
plot(res1d, main="fdr1d and averaged fdr2d")
lines(average.fdr(res2d), col="red")
plot(res2d, grid=TRUE, main="fdr2d is averaged across columns")