CSR

Consensus Clustering for Subclonal structure Reconstruction

Introduction

CSR was originally created for Pan-Cancer Analysis of Whole Genome (PCAWG) working group, Heterogeneity and Evolution, of International Cancer Genome Consortium (ICGC) during the Heterogeneity project. It was used to make a consensus subclonal architecture out of results of 11 participating methods. Please see (Heterogeneity citation) for details.

Prerequisitions

Part of the calculation depends on SPAMS http://spams-devel.gforge.inria.fr/downloads.html. Please follow the instructions on their website to install spams python3 version (prefer Anaconda distribution).

The preprocess script needs R package "dummies" to work. you can simply do

install.packages('dummies')

in R to install the package.

Installing

There is no need to install CSR, it runs just like your regular python script.

Start Your First Example

For your convenience, we supply an example here with 3 methods. https://github.com/kaixiany/CSR/tree/master/Example_input/

Input format

The input files are tab separated, containing 4 columns: chromosome, position, assignment, CP please see https://github.com/kaixiany/CSR/tree/master/Example_input/sample1/method1Input.tsv for an example.

Please prepare one input file for each method for each sample, and put all input of the same sample in a folder. Please separate samples by folder.

Preprocess

We supply a preprocess script CSR_preprocess.R to prepare the actual input files for CSR.

Rscript CSR_preprocess.R inputDir outputDir DownsampleSize FilteringProp

In our example, you can run

Rscript CSR_preprocess.R Example_input/sample1/ preprocessed/samples1/ 5000 0.3

Running CSR

After obtained the preprocessed files, you can now run CSR.py by

python3 CSR.py path_to_preprocessed path_of_results iteration

where iteration can be positive or negative integer. A positive iteration number specifies how many interations should be done when doing matrix decomposition. A negative iteration number specifies the seconds should be run for matrix decomposition. In our example, we can now run

python3 CSR.py preprocessed/samples1/ results/sample1/ -30

This will write results to results/sample1/, and the matrix decomposition will run for 30 sec.

Output file

There should be 3 files in the output:

'mutation_assignments.txt': one column text file, each row indicates the cluster id of the mutation

'mutations_list.txt': one column text file, each raw is a position of the mutation in the format 'chr_position'

'summary_table.txt': three columns text file, each row is a cluster. First column is the cluster id, second column is number of SNVs in the cluster， the third column is the cellular prevalence of the cluster.