| SmoothTtest-class {ClassComparison} | R Documentation |
Implements the smooth t-test for differential expression as developed by Baggely and Coombes.
SmoothTtest(stats, aname = Group One, bname = Group Two, name = paste(aname, vs., bname)) ## S4 method for signature 'SmoothTtest': as.data.frame(x, row.names=NULL, optional=FALSE) ## S4 method for signature 'SmoothTtest': summary(object, ...) ## S4 method for signature 'SmoothTtest, missing': plot(x, folddiff=3, goodflag=2, badch=4, ccl=0, name=x@name, pch='.', xlab='log intensity', ylab='log ratio', ...)
stats |
An object of the TwoGroupStats class. |
aname |
A character string; the name of the second group |
bname |
A character string; the name of the second group |
name |
A character string; the name of this object |
object |
A SmoothTtest object |
x |
A SmoothTtest object |
row.names |
See the base version of as.data.frame.default |
optional |
See the base version of
as.data.frame.default |
folddiff |
A real number; the level of fold difference considered large enough to be indicated in the plots. |
goodflag |
A real number; the level (in standard deviation units) of the smooth t-statistic considered large enough to be indicated in the plot. |
badch |
A real number; the level of variability in single groups
considered large enough to be worrisome. See the multiple
argument to the plot method in the SingleGroup
class. |
ccl |
A list containing objects of the
ColorCoding class. If left at its default
value of zero, colors are chosen automatically. |
pch |
Default plotting character |
xlab |
Label for the x axis |
ylab |
Label for the y axis |
... |
{The usual extra parameters to generic or plotting routines}
In 2001 and 2002, Baggerly and Coombes developed the smooth t-test for
finding differentially expressed genes in microarray data. Along with
many others, they began by log-transforming the data as a reasonable
step in the direction of variance stabilization. They observed,
however, that The gene-by-gene standard deviations still seemed to
vary in a systematic way as a function of the mean log intensity. By
borrowing strenght across genes and using loess to fit
the observed standard deviations as a function of the mean, one
presumably got a better estimate of the true standard deviation.
These smooth estimates are computed for each of two groups of samples being compared. They are then combined (gene-by-gene using the usual univariate formulas) to compute pooled "smooth" estimates of the standard deviation. These smooth estimates are then used in gene-by-gene t-tests.
The interesting question then arises of how to compute and interpret
p-values associated to these individual tests. The liberal
argument asserts that, because smoothing uses data from hundreds
of measurements to estiamte the standard deviation, it can effectively
be treated as "known" in the t-tests, which should thus be compared
against the normal distribution. A conservative argument claims
that the null distribution should still be the t-distribution with the
degrees of freedom determined in the usual way by the number of
samples. The truth probably lies somewhere in between, and is
probably best approximated by some kind of permutation test. In this
implementation, we take the coward's way out and don't provide any of
those alternatives. You have to extract the t-statistics (from the
smooth.t.statistics slot of the object) and compute your own
p-values in your favorite way. If you base the computations on a
theoretical model rather than a permutation test, then the
Bum class provides a convenient way to account for
multiple testing.
In practice, users will first use a data frame and a classification
vector (or an exprSet) to construct an object of the
TwoGroupStats object. Thisn object can then be handed
directly to the SmoothTtest function to perform the smooth
t-test.
one:SingleGroup class
representing a loess smooth of standard deviation as a function of
the mean in the first group of samples.two:SingleGroup class
representing a loess smooth of standard deviation as a function of
the mean in the second group of samples.smooth.t.statistics:fit:x and
y, containing the smooth estimates of the pooled standard deviationdif:avg:aname:bname:name:stats:TwoGroupStats object that was used
to create this object. SingleGroup objects representing the two groups of
samples. The third plot is a scatter plot comparing the means in
the two groups. The fourth plot is Bland-Altman plot of the
overall mean against the difference in means (also known
colloquially as an M-vs-A plot). The fifth plot is a histogram of
the smooth t-statistics. The final plot is a scatter plot of the
smooth t-statistics as a function of the mean intensity. Colors in
the plots are controlled by the curent values of
COLOR.BORING,
COLOR.SIGNIFICANT,
COLOR.BAD.REPLICATE,
COLOR.WORST.REPLICATE,
COLOR.FOLD.DIFFERENCE,
COLOR.CENTRAL.LINE, AND
COLOR.CONFIDENCE.CURVE.
Kevin R. Coombes <kcoombes@mdanderson.org>
Baggerly, K.A., Coombes, K.R., Hess, K.R., Stivers, D.N., Abruzzo, L.V., Zhang, W. Identifying differentially expressed genes in cDNA microarray experiments. J Comp Biol. 8:639-659, 2001.
Coombes, K.R., Highsmith, W.E., Krogmann, T.A., Baggerly, K.A., Stivers, D.N., Abruzzo, L.V. Identifying and quantifying sources of variation in microarray data using high-density cDNA membrane arrays. J Comp Biol. 9:655-669, 2002.
Altman DG, Bland JM. Measurement in Medicine: the Analysis of Method Comparison Studies. The Statistician, 1983; 32: 307-317.
Bum, MultiTtest,
SingleGroup, TwoGroupStats.
bogus <- matrix(rnorm(30*1000, 8, 3), ncol=30, nrow=1000) splitter <- rep(FALSE, 30) splitter[16:30] <- TRUE x <- TwoGroupStats(bogus, splitter) y <- SmoothTtest(x) opar <- par(mfrow=c(2, 3), pch='.') plot(y, badch=2, goodflag=1) par(opar) # cleanup rm(bogus, splitter, x, y, opar)