\documentclass[11pt]{article}

\usepackage{times}
\usepackage{hyperref}
\usepackage[authoryear,round]{natbib}
\usepackage{times}
\usepackage{comment}
\textwidth=6.2in
\textheight=8.5in
%\parskip=.3cm
\oddsidemargin=.1in
\evensidemargin=.1in
%\headheight=-.3in
%\VignetteIndexEntry{UsingPloGO2WithWGCNA}
%\VignetteDepends{xtable}

\bibliographystyle{plainnat}

\title{Integration of PloGO2 and WGCNA for proteomics}
\author{J.Wu and D.Pascovici}

\begin{document}

\maketitle
\begin{abstract}
This vignette describes a tailored workflow of using WGCNA for protein network analysis. A publicly available protein ratio dataset was used to 
demonstrate the major workflow steps and its outputs \cite{YunqiWu2016}. It also demonstrates that the WGCNA output can be seamlessly input into PloGO2 and
further characterised functionally using PloGO2. 
\end{abstract}


\section{WGCNA workflow for proteomics}
\label{sec_wgcna}
WGCNA \cite{langfelder2008wgcna} is a popular correlation network functional analysis tool
which can be used to generate a potentially large number of related data subsets via clustering. 
The WGCNA was proposed and mostly used for genomic data correlation analysis. 
We tailored the WGCNA workflow to be more suitable for proteomic data analysis, and include
this wrapper in the PloGO2 package for completeness.  

Our in-house tailored WGNCA workflow starts from a protein abundance or ratio dataset. 
It is assumed that the data has been 
properly normalised. The protein expression will be log-transformed before the analysis starts. 
Firstly, a soft power was selected for constructing the weighted correlation matrix by using the 
approximate scale-free network criteria.  A scale free network is one where the topology is dominated 
by a few highly connected nodes, which link the rest of the less connected nodes to the system; 
it is assumed that most biologically relevant networks should satisfy this property.  
By raising the correlation to the selected soft power, the correlation network becomes scale-free.  
The cut-off of the parameter RsquaredCut was set as 0.85. Signed networks, instead of unsigned for 
gene network analysis, are constructed using the soft power selected.  
Then TOM (topology overlap metrics) distance was calculated from the network adjacency. 
Hierarchical clustering was performed based on the TOM distance. 
A set of clusters were obtained by using the dynamic tree cutting method with the parameter 
value of minClusterSize as 20. An automatic cluster merging function was invoked to merge 
the closely correlated clusters from the dynamic tree cutting.

The script for an example WGNCA workflow can be found in the script folder of this package.
An example of the input of WGCNA is as follows.

<<wgcnainput, results=tex, eval=TRUE, echo=TRUE>>=
library(PloGO2)
library(xtable)
path <- system.file("files", package = "PloGO2")
print(xtable(read.csv(file.path(path,"rice.csv"))[1:6,c(1:2,5:7)]))
@

The included WGNCA workflow can be executed using the following command
<<wgcnaplots, eval=TRUE, echo=TRUE, fig=FALSE, results=hide>>=
source(file.path(system.file("script", package = "PloGO2"), "WGCNA_proteomics.R"))
@

A number of plots will be produced to visualise the overall weighted correlation network, cluster profiles and 
eigenprotein boxplots. Some examples are as follows.

Fig \ref{fig:clusterdendrogram} shows the cluster dendrogram for the initial and merged clusters.

\begin{figure}[htb!]
\centering
\includegraphics{DendroColorMergedClust.png}
\caption{Cluster dendrogram}
\label{fig:clusterdendrogram}
\end{figure}


Fig \ref{fig:hubboxplot} shows the boxplots for the top 6 hub proteins for the red cluster.

\begin{figure}[htb!]
\centering
\includegraphics[height=9cm, width=9cm]{Boxplot hub proteins - red.png}
\caption{Boxplot for hub proteins - red}
\label{fig:hubboxplot}
\end{figure}


Fig \ref{fig:clusterprofile} shows the cluster profile for all clusters.

\begin{figure}[htb!]
\centering
\includegraphics[height=9cm, width=9cm]{WGCNAClusterPattenME.png}
\caption{Cluster profile}
\label{fig:clusterprofile}
\end{figure}


The output of the WGCNA analysis can be summarised in a multi-tab Excel file which include
the overall data and proteins in every cluster ordered by their kME (fuzzy module membership) values.
Hub proteins will be easily identified The proteins with the highest kMEs can be viewed as the hub proteins of that cluster. 
Therefore, the hub proteins are those on the top of the list. An example of the result for one cluster, "red",
is as follows.

<<wgcnaoutput, results=tex, eval=TRUE, echo=TRUE>>=
print(xtable(readWorkbook("ResultsWGCNA.xlsx", "red")[1:6,c(1:2,5:7,23:24)]))
@

\section{Using input from WGCNA workflow}
The output of the WGCNA can be directly fed into PloGO2 to perform the KEGG pathway analysis, with
pre-processed KEGG pathway DB and optional abundance file (refer to the PloGO2 vignette for details).

<<wgcna2plogo2, eval=TRUE, echo=TRUE>>=
annot.folder <- genAnnotationFiles("ResultsWGCNA.xlsx", DB.name = file.path(path, "pathwayDB.csv"))
res <- PloPathway(reference="AllData", filesPath=annot.folder,
	data.file.name = file.path(path, "Abundance_data.csv"),
	datafile.ignore.cols = 1)
@


\bibliography{REFERENCES}

\end{document}