Molecular subtyping with GISTIC2 results

Prepare as input gistic score for cNMF

transform.gistic.scores.as.cnmf.input <- function(x)
{
  x1 <- x
  x2 <- x
  x1[x1<0] <- 0
  x2[x2>0] <- 0
  rbind(x1, -x2)
}

infile <- 'all_lesions.conf_95.txt'
d <- read.table(infile, header=TRUE,sep="\t", stringsAsFactors=FALSE)

score.idx <- grepl('CN values',d[,1]) == TRUE
status.idx <- grepl('CN values',d[,1]) == FALSE
gistic.scores <- d[score.idx, 10:(ncol(d) - 1)] ## The last column in d is NA
cna.statuses <- d[status.idx, 10:(ncol(d) - 1)] 

score.cyto.names <- paste(paste(sub("\\s+",'', d[score.idx, 2])), 1:(length(score.idx)/2), sep="__")
status.cyto.names <- paste(paste(sub("\\s+",'', d[status.idx, 2])), 1:(length(status.idx)/2), sep="__")
sample.names <- colnames(d)[10:(ncol(d) - 1)]

transformed.gistic.scores <- transform.gistic.scores.as.cnmf.input(gistic.scores)
transformed.score.cyto.names <- c(paste(score.cyto.names, 'Pos', sep="__"), paste(score.cyto.names, "Neg", sep="__"))

## Export preprocessed data to files for cNMF
write.table(transformed.gistic.scores, file='transformed.gistic.scores.txt', sep="\t", quote=FALSE, row.names=FALSE, col.names=FALSE)
write.table(transformed.score.cyto.names, file='transformed.score.cyto.names.txt', sep="\t", quote=FALSE, row.names = FALSE, col.names = FALSE)

write.table(cna.statuses, file='cna.statuses.txt', sep="\t", quote=FALSE, row.names=FALSE, col.names=FALSE)
write.table(status.cyto.names, file='status.cyto.names.txt', sep="\t", quote=FALSE, row.names = FALSE, col.names = FALSE)

write.table(sample.names, file="sample.names.txt", quote=FALSE, row.names = FALSE, col.names = FALSE)

Running cNMF (On BGI server)

#!/bin/bash
rowNameFile='transformed.score.cyto.names.txt'           ## row names must be unique!
##rowNameFile='status.cyto.names.txt'
colNameFile='sample.names.txt'         ## column names must be unique!
inputFile='transformed.gistic.scores.txt'       ## input matrix
##inputFile='cna.statuses.txt'
kstart=2                            ## nmf rank
kend=10                             ## nmf rank
nloop=100                           ## nloop for each rank, e.g. 100, 50 etc.
algorithm=als            # Algorithms: brunet, sNMFR, gdclsNMF, mult, als
transposeInputMatrix=0   # Whether transpose input matrix: 1 - transpose, 0 - do not transpose
verbose=1
maxiter=10000000

MCRroot=/ifshk1/BC_CANCER/03user/lixiangchun/Software/INSTALL/MCR_R2013a/INSTALL/v81
DES_PATH=/ifshk1/BC_CANCER/03user/lixiangchun/iCGA/v0.02/decipherExpressionSignatures

# The proto-type of matlab function is:
# function decipherExpressionSignatures(rowNameFile, colNameFile, inputFile, kstart, kend, nloop, algorithm, transposeInputMatrix, verbose, maxiter)

bash $DES_PATH/run_decipherExpressionSignatures.sh $MCRroot $rowNameFile $colNameFile $inputFile $kstart $kend $nloop $algorithm $transposeInputMatrix $verbose $maxiter

# If use prefer a little bit faster version, try the following:
#bash $DES_PATH/run_cnmfexpression.sh $MCRroot $rowNameFile $colNameFile $inputFile $kstart $kend $nloop $algorithm $transposeInputMatrix $verbose $maxiter

Visualising cNMF results with MATLAB

> load Output.mat;
> consplot(ordcons, coph, rho)

function consplot(cons, coph, rho)

k0 = 2;
% 
% plotting cNMF consensus matrices
% Usage: consplot(ordcons, coph, rho)
%

[k,m,m] = size(cons);
j = 1;
nr = 3;
nc = 4;

if size(cons,1) > nr * nc
    nr = sqrt(size(cons,1) + 3);
    nc = nr;
end

for i = 1:(k-1)
    subplot(nr, nc,i)
    %subplot(size(cons,1)/2,2,i)
    u = cons(k0+i-1,:,:);
    imagesc(squeeze(u));
    title(sprintf('k = %d', k0+i-1));
    j = i;
end

subplot(nr,nc,j+1)
plot(coph, '-o')
xlabel('cNMF rank')
ylabel('Cophenetic coeff.')
%ylim([0.7,1])
xlim([2, size(coph,2)])
grid on

subplot(nr,nc,j+2)
plot(rho, '-o')
xlabel('cNMF rank')
%ylabel('\rho')
%ylabel('\tau')
ylabel('Dispersion coeff.')
%ylim([0.7,1])
xlim([2, size(rho,2)])
grid on