---
title: "QUBIC Tutorial"
author:
- Yu Zhang
- Juan Xie
- Qin Ma
classoption: hyperref,
vignette: >
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteIndexEntry{QUBIC Tutorial}
output:
  pdf_document:
    fig_caption: yes
    pandoc_args: --listings
  html_document:
    highlight: tango
    theme: flatly
  word_document:
    fig_caption: yes
    highlight: tango
    md_extensions: -autolink_bare_uris
header-includes:
- \usepackage{xcolor}
- \lstset{ backgroundcolor=\color[RGB]{248,248,248}, basewidth=0.5em, basicstyle=\small\ttfamily,
  breakatwhitespace=true, breakautoindent=true, breaklines=true, captionpos=b, columns=flexible,
  commentstyle=\color[rgb]{0.56,0.35,0.01}\itshape, escapeinside={\%*}{*)}, frame=tb,
  keepspaces=true, keywordstyle=\color[rgb]{0.13,0.29,0.53}\bfseries, linewidth=\textwidth,
  numberstyle=\footnotesize, showspaces=false, showstringspaces=false, showtabs=false,
  stringstyle=\color[rgb]{0.31,0.60,0.02}, tabsize=2, }
bibliography: bibliography.bib
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
options(width=98)
```

Gene expression data is very important in experimental molecular biology [@brazma00], especially for cancer study [@fehrmann15]. The large-scale microarray data and RNA-seq data provide good opportunity to do the gene co-expression analyses and identify co-expressed gene modules; and the effective and efficient algorithms are needed to implement such analysis. Substantial efforts have been made in this field, such as @cheng00, Plaid [@lazzeroni02], Bayesian Biclustering [BCC, @gu08], among them Cheng and Church and Plaid has the R package implementation. It is worth noting that our in-house biclustering algorithm, QUBIC [@li09], is reviewed as one of the best programs in terms of their prediction performance on benchmark datasets. Most importantly, it is reviewed as the best one for large-scale real biological data [@eren12].

Until now, QUBIC has been cited over 120 times (via [Google Scholar](https://scholar.google.com/scholar?cites=15047943471221701463)) and its web server, QServer, was developed in 2012 to facilitate the users without comprehensive computational background [@zhou12]. In the past five years, the cost of RNA-sequencing decreased dramatically, and the amount of gene expression data keeps increasing. Upon requests from users and collaborators, we developed this R package of QUBIC to void submitting large data to a webserver.

The unique features of our R package [@zhang16] include (1) updated and more stable back-end resource code (re-written by C++), which has better memory control and is more efficient than the one published in 2009. For an input dataset in *Arabidopsis*, with 25,698 genes and 208 samples, we observed more than 40% time saving; and (2) comprehensive functions and examples, including discretize function, heatmap drawing and network analysis. 

# How to cite

```{r, comment=NA, message=FALSE}
citation("QUBIC")
```

# Other languages

If R is not your thing, there is also [a C version of *QUBIC*](https://github.com/maqin2001/QUBIC).

# Help 

If you are having trouble with this R package, contact [the maintainer, Yu Zhang](mailto:zy26@jlu.edu.cn). 

# Install and load

Stable version from BioConductor


```r
source("https://bioconductor.org/biocLite.R")
biocLite("QUBIC")
```

Or development version from GitHub


```r
install.packages("devtools")
devtools::install_github("zy26/QUBIC")
```

Load *QUBIC*


```{r, message=FALSE}
library("QUBIC")
```

Functions
======================

There are nine functions provided by QUBIC package. 

* *qudiscretize()* creates a discrete matrix for a given gene expression matrix;
* *BCQU()* performs a qualitative biclustering for real matrix;
* *BCQUD()* performs a qualitative biclustering for discretized matrix;
* *quheatmap()* can draw heatmap for singe bicluster or overlapped biclusters;
* *qunetwork()* can automatically create co-expression networks based on the identified biclusters by QUBIC;
* *qunet2xml()* can convert the constructed co-expression networks into XGMML format for further network analysis in Cytoscape, Biomax and JNets; 
* *query-based biclustering* allows users to input additional biological information to guide the biclustering progress;
* *bicluster expanding* expands existing biclusters under specified consistency level;
* *biclusters comparison* compares biclusters obtained via different algorithms or parameters.

The following examples illustrate how these functions work.

# Example of a random matrix with two diferent embedded biclusters

```{r random, cahce = TRUE, message = FALSE, fig.cap = 'Heatmap for two overlapped biclusters in the simulated matrix'}
library(QUBIC)
set.seed(1)
# Create a random matrix
test <- matrix(rnorm(10000), 100, 100)
colnames(test) <- paste("cond", 1:100, sep = "_")
rownames(test) <- paste("gene", 1:100, sep = "_")

# Discretization
matrix1 <- test[1:7, 1:4]
matrix1

matrix2 <- qudiscretize(matrix1)
matrix2

# Fill bicluster blocks
t1 <- runif(10, 0.8, 1)
t2 <- runif(10, 0.8, 1) * (-1)
t3 <- runif(10, 0.8, 1) * sample(c(-1, 1), 10, replace = TRUE)
test[11:20, 11:20] <- t(rep(t1, 10) * rnorm(100, 3, 0.3))
test[31:40, 31:40] <- t(rep(t2, 10) * rnorm(100, 3, 0.3))
test[51:60, 51:60] <- t(rep(t3, 10) * rnorm(100, 3, 0.3))

# QUBIC
res <- biclust::biclust(test, method = BCQU())
summary(res)

# Show heatmap
hmcols <- colorRampPalette(rev(c("#D73027", "#FC8D59", "#FEE090", "#FFFFBF", 
    "#E0F3F8", "#91BFDB", "#4575B4")))(100)
# Specify colors

par(mar = c(4, 5, 3, 5) + 0.1)
quheatmap(test, res, number = c(1, 3), col = hmcols, showlabel = TRUE)

```

# Example of *Saccharomyces cerevisiae*

```{r yeast, cache = TRUE}
library(QUBIC)
data(BicatYeast)

# Discretization
matrix1 <- BicatYeast[1:7, 1:4]
matrix1

matrix2 <- qudiscretize(matrix1)
matrix2

# QUBIC
x <- BicatYeast
system.time(res <- biclust::biclust(x, method = BCQU()))

summary(res)
```

We can draw heatmap for single bicluster.

```{r yeast-heatmap, cache = TRUE, message = FALSE, fig.cap = 'Heatmap for the second bicluster identified in the *Saccharomyces cerevisiae* data. The bicluster consists of 53 genes and 5 conditions', fig.show = 'asis'}

# Draw heatmap for the second bicluster identified in Saccharomyces cerevisiae data

library(RColorBrewer)
paleta <- colorRampPalette(rev(brewer.pal(11, "RdYlBu")))(11)
par(mar = c(5, 4, 3, 5) + 0.1, mgp = c(0, 1, 0), cex.lab = 1.1, cex.axis = 0.5, 
    cex.main = 1.1)
quheatmap(x, res, number = 2, showlabel = TRUE, col = paleta)

```

We can draw heatmap for overlapped biclusters.

```{r yeast-heatmap2, cache = TRUE, message = FALSE, fig.cap = 'Heatmap for the second and third biclusters identified in the *Saccharomyces cerevisiae* data. Bicluster #2 (topleft) consists of 53 genes and 5 conditions, and bicluster #3 (bottom right) consists of 37 genes and 7 conditions.', fig.show = 'asis'}

# Draw for the second and third biclusters identified in Saccharomyces cerevisiae data

par(mar = c(5, 5, 5, 5), cex.lab = 1.1, cex.axis = 0.5, cex.main = 1.1)
paleta <- colorRampPalette(rev(brewer.pal(11, "RdYlBu")))(11)
quheatmap(x, res, number = c(2, 3), showlabel = TRUE, col = paleta)

```

We can draw network for single bicluster.

```{r yeast-network1, cache = TRUE, message = FALSE, fig.cap = 'Network for the second bicluster identified in the *Saccharomyces cerevisiae* data.', fig.show = 'asis'}

# Construct the network for the second identified bicluster in Saccharomyces cerevisiae
net <- qunetwork(x, res, number = 2, group = 2, method = "spearman")
if (requireNamespace("qgraph", quietly = TRUE))
    qgraph::qgraph(net[[1]], groups = net[[2]], layout = "spring", minimum = 0.6, 
        color = cbind(rainbow(length(net[[2]]) -  1), "gray"), edge.label = FALSE)

```

We can also draw network for overlapped biclusters.

```{r yeast-network2, cache = TRUE, message = FALSE, fig.cap = 'Network for the second and third biclusters identified in the *Saccharomyces cerevisiae* data.', fig.show = 'asis'}

net <- qunetwork(x, res, number = c(2, 3), group = c(2, 3), method = "spearman")
if (requireNamespace("qgraph", quietly = TRUE))
    qgraph::qgraph(net[[1]], groups = net[[2]], layout = "spring", minimum = 0.6, 
        legend.cex = 0.5, color = c("red", "blue", "gold", "gray"), edge.label = FALSE)

```

```r
# Output overlapping heatmap XML, could be used in other software such
# as Cytoscape, Biomax or JNets
sink('tempnetworkresult.gr')
qunet2xml(net, minimum = 0.6, color = cbind(rainbow(length(net[[2]]) - 1), "gray"))
sink()
# We can use Cytoscape, Biomax or JNets open file named 'tempnetworkresult.gr'
```

# Example of *Escherichia coli* data

The *Escherichia coli* data consists of 4,297 genes and 466 conditions.

```{r ecoli, cache = TRUE}
library(QUBIC)
library(QUBICdata)
data("ecoli", package = "QUBICdata")

# Discretization
matrix1 <- ecoli[1:7, 1:4]
matrix1

matrix2 <- qudiscretize(matrix1)
matrix2

# QUBIC
res <- biclust::biclust(ecoli, method = BCQU(), r = 1, q = 0.06, c = 0.95, o = 100, 
    f = 0.25, k = max(ncol(ecoli)%/%20, 2))
system.time(res <- biclust::biclust(ecoli, method = BCQU(), r = 1, q = 0.06, c = 0.95, 
    o = 100, f = 0.25, k = max(ncol(ecoli)%/%20, 2)))
summary(res)

```

```{r ecoli-heatmap, cache = TRUE, message = FALSE, fig.cap = 'Heatmap for the fifth bicluster identified in the *Escherichia coli* data. The bicluster consists of 103 genes and 38 conditions', fig.show = 'asis'}

# Draw heatmap for the 5th bicluster identified in Escherichia coli data

library(RColorBrewer)
paleta <- colorRampPalette(rev(brewer.pal(11, "RdYlBu")))(11)
par(mar = c(5, 4, 3, 5) + 0.1, mgp = c(0, 1, 0), cex.lab = 1.1, cex.axis = 0.5, 
    cex.main = 1.1)
quheatmap(ecoli, res, number = 5, showlabel = TRUE, col = paleta)

```

```{r ecoli-heatmap2, cache = TRUE, message = FALSE, fig.cap = 'Heatmap for the fourth and eighth biclusters identified in the *Escherichia coli* data. Bicluster #4 (topleft) consists of 108 genes and 44 conditions, and bicluster #8 (bottom right) consists of 26 genes and 33 conditions', fig.show = 'asis'}

library(RColorBrewer)
paleta <- colorRampPalette(rev(brewer.pal(11, "RdYlBu")))(11)
par(mar = c(5, 4, 3, 5), cex.lab = 1.1, cex.axis = 0.5, cex.main = 1.1)
quheatmap(ecoli, res, number = c(4, 8), showlabel = TRUE, col = paleta)

```

```{r ecoli-network, cache = TRUE, message = FALSE, fig.cap = 'Network for the fifth bicluster identified in the *Escherichia coli* data.', fig.show = 'asis' }

# construct the network for the 5th identified bicluster in Escherichia coli data
net <- qunetwork(ecoli, res, number = 5, group = 5, method = "spearman")
if (requireNamespace("qgraph", quietly = TRUE))
    qgraph::qgraph(net[[1]], groups = net[[2]], layout = "spring", minimum = 0.6, 
        color = cbind(rainbow(length(net[[2]]) - 1), "gray"), edge.label = FALSE)


```

```{r ecoli-network2, cache = TRUE, message = FALSE, fig.cap = 'Network for the fourth and eighth biclusters identified in the *Escherichia coli* data.', fig.show = 'asis'}

# construct the network for the 4th and 8th identified bicluster in Escherichia coli data
net <- qunetwork(ecoli, res, number = c(4, 8), group = c(4, 8), method = "spearman")
if (requireNamespace("qgraph", quietly = TRUE))
    qgraph::qgraph(net[[1]], groups = net[[2]], legend.cex = 0.5, layout = "spring", 
        minimum = 0.6, color = c("red", "blue", "gold", "gray"), edge.label = FALSE)

```

## Query-based biclustering

We can conduct a query-based biclustering by adding the weight parameter. In this example, the instance file "511145.protein.links.v10.txt" [@szklarczyk14] was downloaded from string (http://string-db.org/download/protein.links.v10/511145.protein.links.v10.txt.gz) and decompressed and saved in working directory. 

```r
# Here is an example to download and extract the weight
library(igraph)
url <- "http://string-db.org/download/protein.links.v10/511145.protein.links.v10.txt.gz"
tmp <- tempfile()
download.file(url, tmp)
graph = read.graph(gzfile(tmp), format = "ncol")
unlink(tmp)
ecoli.weight <- get.adjacency(graph, attr = "weight")
```


```{r query-based-biclustering, cache = TRUE, message = FALSE}
library(QUBIC)
library(QUBICdata)
data("ecoli", package = "QUBICdata")
data("ecoli.weight", package = "QUBICdata")
res0 <- biclust(ecoli, method = BCQU(), verbose = FALSE)
res0
res4 <- biclust(ecoli, method = BCQU(), weight = ecoli.weight, verbose = FALSE)
res4
```

## Bicluster-expanding

we can expand existing biclustering results to recruite more genes according to certain consistency level:

```{r bicluster-expanding, cache = TRUE}
res5 <- biclust(x = ecoli, method = BCQU(), seedbicluster = res, f = 0.25, verbose = FALSE)
summary(res5)
```

## Biclusters comparison

We can compare the biclustering results obtained from different algorithms, or from a same algorithm with different combinations of parameter.

```{r biclusters-comparison, cache = TRUE}
test <- ecoli[1:50,]
res6 <- biclust(test, method = BCQU(), verbose = FALSE)
res7 <- biclust (test, method = BCCC())
res8 <- biclust(test, method = BCBimax())
showinfo (test, c(res6, res7, res8))
```

# References