boot.null: Non-parametric bootstrap resampling function in package...

Description Usage Arguments Value Note Author(s) References See Also Examples

View source: R/nulldistn_c.R

Description

Given a data set and a closure, which consists of a function for computing the test statistic and its enclosing environment, this function produces a non-parametric bootstrap estimated test statistics null distribution. The observations in the data are resampled using the ordinary non-parametric bootstrap is used to produce an estimated test statistics distribution. This distribution is then transformed to produce the null distribution. Options for transforming the nonparametric bootstrap distribution include center.only, center.scale, and quant.trans. Details are given below. These functions are called by MTP and EBMTP.

Usage

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
boot.null(X, label, stat.closure, W = NULL, B = 1000, test, nulldist, theta0 = 0, tau0 = 1, marg.null = NULL, marg.par = NULL, 
    ncp = 0, perm.mat, alternative = "two.sided", seed = NULL, 
    cluster = 1, dispatch = 0.05, keep.nulldist, keep.rawdist)

boot.resample(X, label, p, n, stat.closure, W, B, test)

center.only(muboot, theta0, alternative)

center.scale(muboot, theta0, tau0, alternative)

quant.trans(muboot, marg.null, marg.par, ncp, alternative, perm.mat) 

Arguments

X

A matrix, data.frame or ExpressionSet containing the raw data. In the case of an ExpressionSet, exprs(X) is the data of interest and pData(X) may contain outcomes and covariates of interest. For boot.resample X must be a matrix. For currently implemented tests, one hypothesis is tested for each row of the data.

label

A vector containing the class labels for t- and F-tests.

stat.closure

A closure for test statistic computation, like those produced internally by the MTP function. The closure consists of a function for computing the test statistic and its enclosing environment, with bindings for relevant additional arguments (such as null values, outcomes, and covariates).

W

A vector or matrix containing non-negative weights to be used in computing the test statistics. If a matrix, W must be the same dimension as X with one weight for each value in X. If a vector, W may contain one weight for each observation (i.e. column) of X or one weight for each variable (i.e. row) of X. In either case, the weights are duplicated appropriately. Weighted F-tests are not available. Default is 'NULL'.

B

The number of bootstrap iterations (i.e. how many resampled data sets) or the number of permutations (if nulldist is 'perm'). Can be reduced to increase the speed of computation, at a cost to precision. Default is 1000.

test

Character string specifying the test statistics to use. See MTP for a list of tests.

theta0

The value used to center the test statistics. For tests based on a form of t-statistics, this should be zero (default). For F-tests, this should be 1.

tau0

The value used to scale the test statistics. For tests based on a form of t-statistics, this should be 1 (default). For F-tests, this should be 2/(K-1), where K is the number of groups. This argument is missing when center.only is chosen for transforming the raw bootstrap test statistics.

marg.null

If nulldist='boot.qt', the marginal null distribution to use for quantile transformation. Can be one of 'normal', 't', 'f' or 'perm'. Default is 'NULL', in which case the marginal null distribution is selected based on choice of test statistics. Defaults explained below. If 'perm', the user must supply a vector or matrix of test statistics corresponding to another marginal null distribution, perhaps one created externally by the user, and possibly referring to empirically derived marginal permutation distributions, although the statistics could represent any suitable choice of marginal null distribution.

marg.par

If nulldist='boot.qt', the parameters defining the marginal null distribution in marg.null to be used for quantile transformation. Default is 'NULL', in which case the values are selected based on choice of test statistics and other available parameters (e.g., sample size, number of groups, etc.). Defaults explained below. User can override defaults, in which case a matrix of marginal null distribution parameters can be accepted. Providing a matrix of values allows the user to perform multiple testing using parameters which may vary with each hypothesis, as may be desired in common-quantile minP procedures. In this way, factors affecting multiple testing procedure performance such as sample size or missingness may be assessed.

ncp

If nulldist='boot.qt', a value for a possible noncentrality parameter to be used during marginal quantile transformation. Default is 'NULL'.

perm.mat

If nulldist='boot.qt' and marg.null='perm', a matrix of user-supplied test statistics from a particular distribution to be used during marginal quantile transformation. The statistics may represent empirically derived marginal permutation values, may be theoretical values, or may represent a sample from some other suitable choice of marginal null distribution.

alternative

Character string indicating the alternative hypotheses, by default 'two.sided'. For one-sided tests, use 'less' or 'greater' for null hypotheses of 'greater than or equal' (i.e. alternative is 'less') and 'less than or equal', respectively.

seed

Integer or vector of integers to be used as argument to set.seed to set the seed for the random number generator for bootstrap resampling. This argument can be used to repeat exactly a test performed with a given seed. If the seed is specified via this argument, the same seed will be returned in the seed slot of the MTP object created. Else a random seed(s) will be generated, used and returned. Vector of integers used to specify seeds for each node in a cluster used to to generate a bootstrap null distribution.

cluster

Integer of 1 or a cluster object created through the package snow. With cluster=1, bootstrap is implemented on single node. Supplying a cluster object results in the bootstrap being implemented in parallel on the provided nodes. This option is only available for the bootstrap procedure.

csnull

DEPRECATED as of multtest v. 2.0.0 given expanded null distribution options. Previously, this argument was an indicator of whether the bootstrap estimated test statistics distribution should be centered and scaled (to produce a null distribution) or not. If csnull=FALSE, the (raw) non-null bootstrap estimated test statistics distribution was returned. If the non-null bootstrap distribution should be returned, this object is now stored in the 'rawdist' slot when keep.rawdist=TRUE.

dispatch

The number or percentage of bootstrap iterations to dispatch at a time to each node of the cluster if a computer cluster is used. If dispatch is a percentage, B*dispatch must be an integer. If dispatch is an integer, then B/dispatch must be an integer. Default is 5 percent.

p

An integer of the number of variables of interest to be tested.

n

An integer of the total number of samples.

muboot

A matrix of bootstrapped test statistics.

keep.nulldist

Logical indicating whether to return the computed bootstrap null distribution, by default 'TRUE'. Not available for nulldist='perm'. Note that this matrix can be quite large.

keep.rawdist

Logical indicating whether to return the computed non-null (raw) bootstrap distribution, by default 'FALSE'. Not available for when using nulldist='perm' or 'ic'. Note that this matrix can become quite large. If one wishes to use subsequent calls to update in which one updates choice of bootstrap null distribution, keep.rawdist must be TRUE. To save on memory, update only requires that one of keep.nulldist or keep.rawdist be 'TRUE'.

Value

A list with the following elements:

rawboot

If keep.rawdist=TRUE, the matrix of non-null, non-transformed bootstrap test statistics. If 'FALSE', an empty matrix with dimension 0-by-0.

muboot

If keep.rawdist=TRUE (default), the matrix of appropriately transformed null test statistics as given by one of center.scale, center.only, or quant.trans. This is the estimated joint test statistics null distribution.

Both list elements rawboot and muboot contain matrices of dimension the number of hypotheses (typically nrow(X)) by the number of bootstrap iterations (B). Each row of muboot is the bootstrap estimated marginal null distribution for a single hypothesis. For boot.null and center.scale, each column of muboot is a centered and scaled resampled vector of test statistics. For boot.null and center.only, each column of muboot is a centered, resampled vector of test statistics.

For boot.null and quant.trans, each column of muboot is a marginal null quantile-transformed resampled vector of test statistics. For each choice of marginal null distribution (defined by marg.null and marg.par), a random sample of size B is drawn and then rearranged based on the ranks of the marginal test statistics bootstrap distribution corresponding to each hypothesis (typically within rows of X). This means that using quant.trans will set the RNG seed ahead by B * the number of hypotheses (similarly, typically nrow(X)). Tie breaks in the marginal non-null bootstrap distribution are implemented inside the internal function marg.samp called by quant.trans. Default values of marg.null and marg.par are available based on choice of test statistics, sample size 'n', and various other parameters. By the time boot.null is called in either the MTP or EBMTP functions, the default marginal null distribution settings have already been formatted and passed in their correct form to boot.null. These default values correspond to:

t.onesamp:

t-distribution with df=n-1;

t.twosamp.equalvar:

t-distribution with df=n-2;

t.twosamp.unequalvar:

N(0,1);

t.pair:

t-distribution with df=n-1, where n is the number of unique samples, i.e., the number of observed differences/paired samples;

f:

F-distribution with df1=k-1, df2=n-k, for k groups;

f.block:

NA. Only available with permutation distribution;

f.twoway:

F-distribution with df1=k-1,df2=n-k*l, for k groups and l blocks;

lm.XvsZ:

N(0,1);

lm.YvsXZ:

N(0,1);

coxph.YvsXZ:

N(0,1);

t.cor

t-distribution with df=n-2;

z.cor

N(0,1).

The above defaults, however, can be overridden by manually setting values of marg.null and marg.par.

The rawboot and muboot objects are returned in the slots rawdist and nulldist of an object of class MTP or EBMTP when the arguments keep.rawdist or keep.nulldist to the MTP function are TRUE. For boot.resample a matrix of bootstrap samples prior to null transformation is returned.

Note

Thank you to Duncan Temple Lang and Peter Dimitrov for suggestions about the code.

Author(s)

Katherine S. Pollard, Houston N. Gilbert, and Sandra Taylor, with design contributions from Sandrine Dudoit and Mark J. van der Laan.

References

M.J. van der Laan, S. Dudoit, K.S. Pollard (2004), Augmentation Procedures for Control of the Generalized Family-Wise Error Rate and Tail Probabilities for the Proportion of False Positives, Statistical Applications in Genetics and Molecular Biology, 3(1). http://www.bepress.com/sagmb/vol3/iss1/art15/

M.J. van der Laan, S. Dudoit, K.S. Pollard (2004), Multiple Testing. Part II. Step-Down Procedures for Control of the Family-Wise Error Rate, Statistical Applications in Genetics and Molecular Biology, 3(1). http://www.bepress.com/sagmb/vol3/iss1/art14/

S. Dudoit, M.J. van der Laan, K.S. Pollard (2004), Multiple Testing. Part I. Single-Step Procedures for Control of General Type I Error Rates, Statistical Applications in Genetics and Molecular Biology, 3(1). http://www.bepress.com/sagmb/vol3/iss1/art13/

Katherine S. Pollard and Mark J. van der Laan, "Resampling-based Multiple Testing: Asymptotic Control of Type I Error and Applications to Gene Expression Data" (June 24, 2003). U.C. Berkeley Division of Biostatistics Working Paper Series. Working Paper 121. http://www.bepress.com/ucbbiostat/paper121

M.J. van der Laan and A.E. Hubbard (2006), Quantile-function Based Null Distributions in Resampling Based Multiple Testing, Statistical Applications in Genetics and Molecular Biology, 5(1). http://www.bepress.com/sagmb/vol5/iss1/art14/

S. Dudoit and M.J. van der Laan. Multiple Testing Procedures and Applications to Genomics. Springer Series in Statistics. Springer, New York, 2008.

See Also

corr.null, MTP, MTP-class, EBMTP, EBMTP-class, get.Tn, ss.maxT, mt.sample.teststat,get.Tn, wapply, boot.resample

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
set.seed(99)
data<-matrix(rnorm(90),nr=9)

#closure
ttest<-meanX(psi0=0,na.rm=TRUE,standardize=TRUE,alternative="two.sided",robust=FALSE)

#test statistics
obs<-get.Tn(X=data,stat.closure=ttest,W=NULL)

#bootstrap null distribution (B=100 for speed, default nulldist, "boot.cs")
nulldistn<-boot.null(X=data,W=NULL,stat.closure=ttest,B=100,test="t.onesamp",
	nulldist="boot.cs",theta0=0,tau0=1,alternative="two.sided",
	keep.nulldist=TRUE,keep.rawdist=FALSE)$muboot

#bootstrap null distribution with marginal quantile transformation showing
#default values that are passed to marg.null and marg.par arguments
nulldistn.qt<-boot.null(X=data,W=NULL,stat.closure=ttest,B=100,test="t.onesamp",
	nulldist="boot.qt",theta0=0,tau0=1,alternative="two.sided",
	keep.nulldist=TRUE,keep.rawdist=FALSE,marg.null="t",
	marg.par=matrix(9,nr=10,nc=1))$muboot

#unadjusted p-values
rawp<-apply((obs[1,]/obs[2,])<=nulldistn,1,mean)
sum(rawp<=0.01)

rawp.qt<-apply((obs[1,]/obs[2,])<=nulldistn.qt,1,mean)
sum(rawp.qt<=0.01)

multtest documentation built on Nov. 8, 2020, 11:03 p.m.

Related to boot.null in multtest...