\documentclass[a4paper]{article}
\usepackage[round]{natbib}
\usepackage{Sweave}
\usepackage{hyperref}
%% need no \usepackage{Sweave.sty}
%\usepackage{natbib}
%\usepackage{/usr/lib/R/share/texmf/Sweave}
%\VignetteIndexEntry{Cancer Classification Using Mass Spectrometry-based Proteomics Data}
%\VignetteDepends{pROC}
%\SweaveOpts{keep.source=FALSE}
\hypersetup{%
  pdftitle = {Cancer Classification Using Mass Spectrometry-based Proteomics Data},
  pdfsubject = {package vignette},
  pdfauthor = {Zhu Wang},
%% change colorlinks to false for pretty printing
  colorlinks = {true},
  linkcolor = {blue},
  citecolor = {blue},
  urlcolor = {red},
  hyperindex = {true},
  linktocpage = {true},
}

\author{Zhu Wang \\
  University of Tennessee Health Science Center\\
  zwang145@uthsc.edu}
\title{Cancer Classification Using Mass Spectrometry-based Proteomics Data}
\begin{document}
\date{}
\maketitle


This document presents data analysis similar to \cite{wang2011hingeboost} using R package \textbf{bst}. Serum samples were collected from 77 benign prostate hyperplasia (BPH), 84 early stage prostate cancer, 83 late stage prostate cancer and 82 age-matched healthy men (HM). Peak detection and alignment were based on surface-enhanced laser desorption/ionization (SELDI) mass spectrometry protein profiles. The data have 779 peaks (predictors) for each sample. 
<<echo=false,results=hide>>=
options(prompt = "R> ", continue = " ", width = 70, digits =4, useFancyQuotes = FALSE)
@
<<echo=TRUE, results=hide>>=
library("bst")
library("pROC")
### benign prostate hyperplasia
bph <- read.table(file=system.file("extdata", "BPH.txt", package="bst"))
### early stage prostate cancer, 
ccd <- read.table(file=system.file("extdata", "CCD.txt", package="bst"))
### late stage prostate cancer
cab <- read.table(file=system.file("extdata", "CAB.txt", package="bst"))
### age-matched healthy men
hm <- read.table(file=system.file("extdata", "control.txt", package="bst"))
### mass spectrometry protein
mz <- read.table(file=system.file("extdata", "prostate_mz.txt", package="bst"))
@
\section{Cancer vs Non-cancer}
Cancer includes early and late stage cancer while non-cancer includes healthy men and benign cancer.
<<echo=TRUE>>=
dat <- t(cbind(hm, bph, cab, ccd))
y <- c(rep(-1, dim(hm)[2] + dim(bph)[2]), rep(1,  dim(cab)[2] + dim(ccd)[2]))
@
Peaks with variations only in a very small number of samples are less likely to be linked with
cancer stages. Thus, we filter out the peaks whose total number of fixed value exceed 95\% across samples.
<<>>=
myuniq <- function(x){
  if(sum(x == 0) <= length(x) * 0.95)
  return(TRUE)
  else return(FALSE)
}
res <- apply(dat, 2, myuniq)
dat <- dat[,res]
rownames(dat) <- NULL
colnames(dat) <- mz[res,1]
@
we randomly select 75\% of the samples as
training data and the remaining samples as the test data.
<<>>=
ntrain <- floor(length(y)*0.75)
set.seed(13)
q <- sample(length(y))
y.tr <- y[q][1:ntrain]; y.te <- y[q][-(1:ntrain)]
X <- dat[q,]
x.tr <- X[1:ntrain,];  x.te <- X[-(1:ntrain),] 
@
Apply HingeBoost to classify cancer vs non-cancer.
<<echo=TRUE>>=
dat.m1 <- bst(x=x.tr, y=y.tr, ctrl = bst_control(mstop=400), family = "hinge2")
pred <- predict(dat.m1, x.te)
### misclassification error
mean(abs(y.te-sign(pred))/2)
### area under the curve (AUC) of receiver operating characteristic (ROC)
auc(y.te, pred)
### number of variables selected
length(dat.m1$xselect)
@
Apply twin HingeBoost to classify cancer vs non-cancer.
<<echo=TRUE>>=
dat.m2 <- bst(x=x.tr, y=y.tr, family="hinge2", ctrl = bst_control(mstop=500, 
twinboost=TRUE, twintype=2, coefir=coef(dat.m1), f.init=predict(dat.m1), 
xselect.init = dat.m1$xselect))
pred <- predict(dat.m2, x.te)
### misclassification error
mean(abs(y.te-sign(pred))/2)
### AUC with twin boosting
auc(y.te, pred)
### number of variables selected
length(dat.m2$xselect)
@
\section{Cancer vs Healthy Men}
<<>>=
dat <- t(cbind(cab, ccd, hm))
y <- c(rep(1, dim(cab)[2] + dim(ccd)[2]), rep(-1, dim(hm)[2]))
res <- apply(dat, 2, myuniq)
dat <- dat[,res]
rownames(dat) <- NULL
colnames(dat) <- mz[res,1]
ntrain <- floor(length(y)*0.75)
set.seed(13)
q <- sample(length(y))
y.tr <- y[q][1:ntrain]; y.te <- y[q][-(1:ntrain)]
X <- dat[q,]
x.tr <- X[1:ntrain,];  x.te <- X[-(1:ntrain),] 
@
Apply HingeBoost to classify cancer vs healthy men.
<<echo=TRUE>>=
dat.m1 <- bst(x=x.tr, y=y.tr, ctrl = bst_control(mstop=400), family = "hinge2")
pred <- predict(dat.m1, x.te)
### misclassification error
mean(abs(y.te-sign(pred))/2)
### AUC
auc(y.te, pred)
### number of variables selected
length(dat.m1$xselect)
@
Apply twin HingeBoost to classify cancer vs healthy men.
<<echo=TRUE>>=
dat.m2 <- bst(x=x.tr, y=y.tr, family="hinge2", ctrl = bst_control(mstop=200, 
twinboost=TRUE, twintype=2, coefir=coef(dat.m1), f.init=predict(dat.m1), 
xselect.init = dat.m1$xselect))
pred <- predict(dat.m2, x.te)
### misclassification error
mean(abs(y.te-sign(pred))/2)
### AUC with twin boosting
auc(y.te, pred)
### number of variables selected
length(dat.m2$xselect)
@
\section{Cancer vs Benign Cancer}
<<>>=
dat <- t(cbind(bph, cab, ccd))
y <- c(rep(-1, dim(bph)[2]), rep(1,  dim(cab)[2] + dim(ccd)[2]))
res <- apply(dat, 2, myuniq)
dat <- dat[,res]
rownames(dat) <- NULL
colnames(dat) <- mz[res,1]
ntrain <- floor(length(y)*0.75)
set.seed(13)
q <- sample(length(y))
y.tr <- y[q][1:ntrain]; y.te <- y[q][-(1:ntrain)]
X <- dat[q,]
x.tr <- X[1:ntrain,];  x.te <- X[-(1:ntrain),] 
@
Apply HingeBoost to classify cancer vs benign cancer.
<<echo=TRUE>>=
dat.m1 <- bst(x=x.tr, y=y.tr, ctrl = bst_control(mstop=400), family = "hinge2")
pred <- predict(dat.m1, x.te)
### misclassification error
mean(abs(y.te-sign(pred))/2)
### AUC
auc(y.te, pred)
### number of variables selected
length(dat.m1$xselect)
@
Apply twin HingeBoost to classify cancer vs benign cancer. 
<<echo=TRUE>>=
dat.m2 <- bst(x=x.tr, y=y.tr, family="hinge2", ctrl = bst_control(mstop=500, 
twinboost=TRUE, twintype=2, coefir=coef(dat.m1), f.init=predict(dat.m1), 
xselect.init = dat.m1$xselect))
pred <- predict(dat.m2, x.te)
### misclassification error
mean(abs(y.te-sign(pred))/2)
### AUC with twin boosting
auc(y.te, pred)
### number of variables selected
length(dat.m2$xselect)
@
\section{Benign Cancer vs Healthy Men}
<<>>=
dat <- t(cbind(bph, hm))
y <- c(rep(1, dim(bph)[2]), rep(-1, dim(hm)[2]))
res <- apply(dat, 2, myuniq)
dat <- dat[,res]
rownames(dat) <- NULL
colnames(dat) <- mz[res,1]
ntrain <- floor(length(y)*0.75)
set.seed(13)
q <- sample(length(y))
y.tr <- y[q][1:ntrain]; y.te <- y[q][-(1:ntrain)]
X <- dat[q,]
x.tr <- X[1:ntrain,];  x.te <- X[-(1:ntrain),] 
@
Apply HingeBoost to classify benign cancer vs healthy men.
<<echo=TRUE>>=
dat.m1 <- bst(x=x.tr, y=y.tr, ctrl = bst_control(mstop=400), family = "hinge2")
pred <- predict(dat.m1, x.te)
### misclassification error
mean(abs(y.te-sign(pred))/2)
### AUC
auc(y.te, pred)
### number of variables selected
length(dat.m1$xselect)
@
Apply twin HingeBoost to classify benign cancer vs healthy men.
<<echo=TRUE>>=
dat.m2 <- bst(x=x.tr, y=y.tr, family="hinge2", ctrl = bst_control(mstop=500, 
twinboost=TRUE, twintype=2, coefir=coef(dat.m1), f.init=predict(dat.m1), 
xselect.init = dat.m1$xselect))
pred <- predict(dat.m2, x.te)
### misclassification error
mean(abs(y.te-sign(pred))/2)
### AUC with twin boosting
auc(y.te, pred)
### number of variables selected
length(dat.m2$xselect)
@
%<<sessionInfo>>=
%sessionInfo();
%@
\bibliographystyle{plainnat}
\bibliography{bst}
\end{document}