| corrected.p {swamp} | R Documentation |
Correction of p-values for associations between features and sample annotation
Description
The function corrects for multiple testing of associations of features to sample annotation. Adjustment is done by padjust() or by the p-values from permuted data.
Usage
corrected.p(feature.assoc, correction = "fdr", adjust.permute = T,
adjust.rank = T, ties.method = "first")
Arguments
feature.assoc |
a list with the p-values of feature associations, typically created by the function feature.assoc(). (If not created by feature.assoc() the list has to contain the elements observed.p and permuted.p.) |
correction |
adjustment method to use for padjust(). default="fdr". must be one of "holm", "hochberg", "hommel", "bonferroni", "BH", "BY", "fdr" or "none". |
adjust.permute |
if set to TRUE (default), the p-values will be adjusted by observed.p divided by permuted.p for each rank in observed.p and permuted.p. |
adjust.rank |
if set to TRUE (default), the p-values will be adjusted by calculating for every observed p-value the proportion of smaller permuted.p values to smaller observed.p values. |
ties.method |
if adjust.permute=TRUE or adjust.rank=TRUE the method for handling ties can be either "first" or "random". Tied p-values are likely when using "AUC" as method to measure feature associations. default="first". |
Details
As high-dimensional data contains many features, the p-values of feature associations have to be corrected for multiple testing. The number of features that are significantly associated with sample annotation can show how strog the data is connected to the respective sample annotation. The p-values can be adjusted using the standard correction methods of padjust(). Additionally two methods that use the p-values from permuted data are proposed. First, p-values are adjusted by observed.p divided by permuted.p for each rank in observed.p and permuted.p. For instance if the third lowest p-value in the observed associations is 1e-9 and the third lowest p value in the permuted data is 1e-4, this p-value is correced by 1e-9 divided by 1e-4 which is 1e-5. Second, p-values are adjusted by calculating for every observed p-value the proportion of smaller permuted. p values to smaller observed.p values. For instance, we have a p-value of 1e-3 which is ranked as the 300-lowest p-value in the observed data. In the permuted data there are 3 p-values that are lower than 1e-3. In the 300 p-values we suspect 3 of them to occur by chance, hence the adjusted p-value is 3 divided by 300 which is 0.01. This correction method can yield p-values of 0 and is less robust when only a few permuted.p are smaller than the observed.p. Both proposed correction methods may acutally show similar results to padjust(observed.p,method="fdr")
Value
a list with components
padjust |
a numeric vector containing the corrected p-values using padjust(). |
adjust.permute |
a numeric vector containing the corrected p-values using observed.p divided by permuted.p at each rank |
adjust.rank |
a numeric vector containing the corrected p-values by calculating for every observed p-value the proportion of smaller permuted.p values to smaller observed.p values |
Author(s)
Martin Lauss
Examples
# data as a matrix
set.seed(100)
g<-matrix(nrow=1000,ncol=50,rnorm(1000*50),dimnames=list(paste("Feature",1:1000),
paste("Sample",1:50)))
g[1:100,26:50]<-g[1:100,26:50]+1 # the first 100 features show
# higher values in the samples 26:50
# patient annotations as a data.frame, annotations should be numbers and factor
# but not characters.
# rownames have to be the same as colnames of the data matrix
set.seed(200)
o<-data.frame(Factor1=factor(c(rep("A",25),rep("B",25))),
Factor2=factor(rep(c("A","B"),25)),
Numeric1=rnorm(50),row.names=colnames(g))
#calculate feature associations
res4a<-feature.assoc(g,o$Factor1,method="correlation")
#correct the p-values
res5<-corrected.p(res4a)
names(which(res5$padjust<0.05))
names(which(res5$adjust.permute<0.05))
names(which(res5$adjust.rank<0.05))