Analysis and report generation for ovarian subtyping project

(this will be updated shortly to reflect new directory structure)

Instructions:

First, install MetaGxOvarian (contact Greg Chen or Deena Gendoo for the latest version). Install MetaGx with the commands

git clone https://github.com/bhklab/MetaGx.git R CMD BUILD MetaGx R CMD INSTALL MetaGx_0.9.9.tar.gz

Clone this repository:

git clone https://github.com/bhklab/OvcSubtypes.git

In the "reports" directory, there are three main knitr files: - reproduceResults.Rnw - classificationAcrossDatasets.Rnw - robustness.Rnw Note that classificationAcrossDatasets.Rnw produces the file "esets.not.rescaled.RData" which contains samples of high-grade serious ovarian cancer, with genes not z-score rescaled by gene. In order to ensure consistency in all analyses, this file is used by robustness.Rnw and batch.cluster.all.R.

The three knitr files can be run to produce pdf files within an R session with the command: library(knitr) knit("reproduceResults.Rnw") knit("classificationAcrossDatasets.Rnw") knit("robustness.Rnw") (alternatively, they can be compiled in RStudio, making sure Preferences -> Sweave -> Weave Rnw files is set to knitr)

Note that several dependencies may need to be installed. Packages on CRAN can be installed with, for example, install.packages("xtable") Packages on Bioconductor can be installed with, for example, source("https://bioconductor.org/biocLite.R") biocLite("survcomp")

Within robustness.Rnw, there are two main components: - a reproduction of clustering algorithms and comparison to the cluster labels given in original supplementary texts (producing heatmaps demonstrating concordance between our implementation and the original results) - evaluation of robustness using prediction strength

The prediction strength analysis depends on the output of a fairly computationally intensive run, which produced the output directory jan20clusters. In order to read this output, cd into the "reports" directory and extract the compressed directory: tar -xvfz jan20clusters.tar.gz

(Addendum) Performing cluster analysis for prediction strength analysis:

In the prediction strength analysis, we re-cluster each dataset with each algorithm 100 times. This takes a substantial amount of computation, so this is parallelized on an SGE cluster with the files "submitbatchclustering.sh" and "batch.cluster.all.R". Ensuring that "esets.not.rescaled.RData" exists, the command ./submit_batch_clustering.sh submits an array job to the cluster, creating a directory with clustering output. This directory is read in "robustness_validation.Rnw".

This was most recently performed on January 20, producing the output jan20clusters.tar.gz