Clomial.iterate: Runs EM iterations until convergence of the Clomial model.

Description

Given the data and the initial values for the model parameters, runs EM iterations until convergence of the Clomial model.

Usage

Clomial.iterate(Dt, Dc, Mu, P, maxIt=100, U = NULL, PTrue = NULL,
llCutoff = 10^(-3), computePFunction = compute.P.reparam,
doSilentOptim = TRUE, doTalk = TRUE, doLog = TRUE, debug = FALSE,
noiseReductionRate = 0.01, fliProb=0.05,conservative=TRUE)

Arguments

maxIt

The maximum number of EM iterations.

A matrix which contains the counts of the alternative allele where rows correspond to the genomic loci, and columns correspond to the samples.

A matrix which contains the counts of the total number of mapped reads where rows correspond to the genomic loci, and columns correspond to the samples.

The initial value for the Mu matrix which models the genotypes, where rows and columns correspond to genomic loci and clones, accordingly.

The initial matrix of clonal frequency where rows and columns correspond to clones and samples, accordingly.

The true value for Mu, used for debugging purposes only.

PTrue

The true value for P, used for debugging purposes only.

llCutoff

EM iterations stops if the relative improvement in the log-likelihood is not more than this threshold.

computePFunction

The function used for updating P. For advanced development use only.

doSilentOptim

If TRUE, the optimization massages will not be reported.

doTalk

If FALSE, the function will be run in silent mode.

doLog

Highly recommended to set to TRUE. Then, the computations will be done in log space to avoid numerical issues.

debug

If TRUE, the debug mode will be turned on.

noiseReductionRate

The noise will be reduce by this rate after each EM iteration.

fliProb

A "flipping probability" used for noise injection which can be disabled when fliProb=0. After the first EM iteration, each entry of the matrix Mu such as m may change to 1-m with this probability. This probability decreases on subsequent iterations.

conservative

Boolean where TRUE means noise will be injected only if likelihood is improved after an EM iteration, otherwise the original Mu matrix will be used for the next iteration. For expert use only.

Value

Qs: The history of matrices containing the posterior Q values.
Ps: The history of P matrices.
Mus: The history of Mu matrices.
Mu: The value of Mu after convergence.
P: The value of P after convergence.
llCutoff: The threshold used to decide convergence.
LRatio: The final relative improvement in the log likelihood which lead to convergence.
Likelihoods: The history of log-likelihoods.
fliProb: The final value of fliProb used for noise injection.
timeTaken: An object of class ``difftime'' which reports the total computational time for EM iterations.
endTaken: An object of class ``POSIXct'' (see DateTimeClasses) which reports the time EM iterations finished.

Details

Injecting noise can be done by assigning a positive value to fliProb, and can be disabled by fliProb=0. Noise injection is recommended for training models with a high number of clones (>4).

References

Inferring clonal composition from multiple sections of a breast cancer, Zare et al., Submitted.

Examples

Run this code

set.seed(1)
## Getting data:
data(breastCancer)
Dc <- breastCancer$Dc
Dt <- breastCancer$Dt
freq1 <- Dt/Dc
N <- nrow(Dc)
S <- ncol(Dc)
Cnum <- 4 ## assumed number of clones.
## Random initialization:
random1 <- runif(n=N*(Cnum-1),min=rowMins(freq1)*0.9,max=rowMaxs(freq1)*1.1)
random1[random1>1] <- 1
random1[random1<0] <- 0
Mu <- matrix(random1,N,Cnum-1)
Mu <- cbind( matrix(0,N,1), Mu )
rownames(Mu) <- rownames(Dc)
colnames(Mu) <- paste("C",1:Cnum,sep="")
P <- matrix(runif(Cnum*S),Cnum,S)
rownames(P) <- colnames(Mu)
colnames(P) <- colnames(Dc)
## Normalizing P:
for( t in 1:S ){
	s <- sum(P[,t])
	P[,t] <- P[,t]/s
}##End for.

## Running EM: 
model1 <- Clomial.iterate(Dt=Dt, Dc=Dc, Mu=Mu, P=P)
print("Genotypes:")
round(model1$Mu)
print("Clone frequencies:")
model1$P

Run the code above in your browser using DataLab