PrOType

The Predictor of Ovarian Cancer Type

This repository contains all the data and scripts used to run the Vancouver analysis of the molecular subtypes of High Grade Serous Carcinoma.

Introduction

This documentation will guide you through the process of running the full pipeline, which includes the following components:

1. Unsupervised Learning: use consensus clustering to determine HGSC subtypes using an unsupervised ensemble method
2. Supervised Learning: uses cluster assignments from previous step to classify new samples and learn the best supervised algorithm
3. Post Processing: validates top performing classifiers for different batch correction methods (XPN and CBT), and selects the candidate model for predicting NanoString samples based on validating on overlap data
4. NanoString Classifier: uses candidate model to predict full NanoString data and output predicted subtypes
5. Gene Selection: select top genes based on training batch 1 of NanoString data and make final model predictions
6. Cross Platform: cross platform comparisons and evaluations

Please follow the manual below to run the pipeline on your system. Data can be provided upon request in the form of a zip file. After unzipping the data, place the directory in the repository root and name it data/.

Installation

Docker

PrOType can be run in a preconfigured Docker container. This will allow for a reproducible environment across different machines and operating systems. To install docker on your machine see the official docs: https://docs.docker.com/install/

Once Docker is installed, navigate to the PrOType project directory on your machine then run

docker image build -f Dockerfile -t protype:latest .

to build the image on your machine (this may take some time). After the image has been built, you can start a container with

docker run -it --rm protype:latest bash

Inside the docker container, run git clone https://github.com/AlineTalhouk/PrOType.git and navigate to the PrOType directory. The Dockerfile is already configured with all the dependencies and tools you'll need to run PrOType. For more information on running docker containers, see the official docs: https://docs.docker.com/engine/reference/run/

Packrat

We use packrat as a package management system to keep track of precise versions of all R package dependencies used in PrOType. After you have cloned PrOType to your local machine and open PrOType.Rproj, the packrat bootstrap installation will initialize. Run

packrat::restore()

to restore all packages listed in packrat/packrat.lock from their package sources in packrat/src/ (this will also take some time).

The easiest way to run PrOType on a Linux server is to use git or rsync to copy the project to the server location, and then run the various make targets at the project root. This ensures that packages loaded from bash calls to Rscript use the packrat private library instead of the system library. To share the pipeline with a collaborator, send them the output of

packrat::bundle()

Software Requirements

• Linux environment (distribution CentOS 5.4 was used)
• R version 3.5.1
• git
• GNU make

Usage

Most of the PrOType pipeline is run using a Sun Grid Engine by distributing batch jobs to worker nodes via qsub, thus employing a parallel computing environment. To run the full pipeline, run make all. Separate sections of the pipeline can be run by running their respective make targets. See Makefile for all target names. The user is encouraged to run make in a terminal multiplexer such as screen or tmux so that computationally expensive operations can continue to run in an embedded server and not be killed by server timeouts.

The pipeline is broken into the following analyses:

• All Array Analysis
  • Unsupervised clustering using diceR
  • Supervised classification using splendid
  • Post-Processing
• NanoString Analysis
  • Full prediction of NanoString data
• Gene Selection
  • Reduced gene list
  • Predict final model
• Cross Platform

For a more detailed description of the pipeline, see here.

To use alternative options than the default pipeline parameters, modify the relevant fields in Parameters.sh. Read more about the various parameters here.

Additional Resources

A tree visualization of the file structure can be found here. Helper functions, scripts for job submission, cleaning outputs, and additional developer parameters are found in the assets/ directory.

A Web Tool of the PrOType pipeline is hosted at https://ovcare.shinyapps.io/PrOType. This shiny app allows you to visualize quality control metrics, view normalized data, and predict new NanoString samples using our final model.