DriverNetPy - A Python implementation of DriverNet.

This repository is a Python implementation of the DriverNet algorithm.

Motivation

DriverNet uses an easy-to-understand greedy searching algorithm to prioritize cancer drivers. Here is the paper.

Bashashati A, Haffari G, Ding J, Ha G, Lui K, Rosner J, Huntsman DG, Caldas C, Aparicio SA, Shah SP. 2012. DriverNet: uncovering the impact of somatic driver mutations on transcriptional networks in cancer. Genome Biology. 13:R124. doi:10.1186/gb-2012-13-12-r124.

I like this idea. It is elegant and effective.

To better understand it, I decided to write a Python version in case I would use this algorithm in other projects.

Also, in reading DriverNet paper and its R code. I found that the R implementation of DriverNet has some pitfalls:

In "finding drivers" step, mutated gene candidates were updated by genes already covered by genes selected (line 72 in DriverSelection.R ). This was not mentioned in the paper. I did not know why.
In "significance test" step, the patient-outlier matrix and patient-mutation matrix were shuffled only by gene names (line 35-38 in SignificanceTest.R ), which would keep the number of outliers and mutated genes of each patient unchanged. This is different from the way described in the paper. I think the perturbation described in the paper may be more reasonable.
In "significance test" step, the random patient-outlier matrix and random patient-mutation matrix before running .greedyGeneDriverSelection were preprocessed by a different way as that used in "finding drivers". I did not know why.
In "significance test" step, DriverNet does not use Benjamini-Hochberg approach to correct p-values as stated in the paper.

In addition, DriverNetPy extends the original R version of DriverNet by:

using network permutation scheme as default.
adding parallelization support to run the permutation test faster.

Usage

Prerequisite

DriverNetPy imports the following packages.

Install

git clone git@github.com:YiweiNiu/DriverNetPy.git
cd DriverNetPy

Use

The program can be provoked by DriverNet.py script.

$ Python DriverNetPy/DriverNet.py -h
usage:

python DriverNet.py [-h] <-n network file> <-m mutation file> <-e expression outlier file> [-p permutation_time] [-t threads] [-o output prefix]

optional arguments:
  -h, --help            show this help message and exit
  -n , --network        Path to gene-gene interaction network. (default: None)
  -m , --mutation       Path to sample-mutation file. (default: None)
  -e , --expression     Path to sample-expression file. (default: None)
  -p , --permutation    Permutation times. (default: 500)
  -pg, --not-purturbGraph
                        Not permutation graphs. (default: False)
  -pd, --purturbData    Permutation data. (default: False)
  -t , --threads        Number of CPUs used when permutation. Using all CPUs
                        by default. (default: None)
  -o , --output         Output prefix. (default: DriverNetPy_res)

A Python implementation of DriverNet. See
https://github.com/YiweiNiu/DriverNetPy for details.

Example

Export sample data from the DriverNet R package. These files can be found under the test folder.

library(DriverNet)

data(samplePatientMutationMatrix)
data(samplePatientOutlierMatrix)
data(sampleInfluenceGraph)
data(sampleGeneNames)

# export sample data
write.table(sampleInfluenceGraph, file = 'sampleInfluenceGraph.txt', sep=',', quote = F)
write.table(samplePatientMutationMatrix, file = 'samplePatientMutationMatrix.txt', sep=',', quote = F)
write.table(samplePatientOutlierMatrix, file = 'samplePatientOutlierMatrix.txt', sep=',', quote = F)

# run DriverNet
driversList = computeDrivers(samplePatientMutationMatrix, samplePatientOutlierMatrix,
                             sampleInfluenceGraph, outputFolder=NULL, printToConsole=FALSE)
randomDriversResult = computeRandomizedResult(patMutMatrix=samplePatientMutationMatrix,
                                              patOutMatrix=samplePatientOutlierMatrix,
                                              influenceGraph=sampleInfluenceGraph,
                                              geneNameList= sampleGeneNames, outputFolder=NULL,
                                              printToConsole=FALSE, numberOfRandomTests=20,
                                              weight=FALSE, purturbGraph=FALSE, purturbData=TRUE)
res = resultSummary(driversList, randomDriversResult,
                    samplePatientMutationMatrix, sampleInfluenceGraph,
                    outputFolder=NULL, printToConsole=FALSE)
write.table(res, file='DriverNet_res.txt', quote=F, sep='\t')

Run DriverNetPy using the example data.

$ cd test
$ python ../DriverNetPy/DriverNet.py -n sampleInfluenceGraph.txt -m samplePatientMutationMatrix.txt -e samplePatientOutlierMatrix.txt
2019-05-15 16:45:13,177 - __main__ - INFO - Start running.
2019-05-15 16:45:13,177 - __main__ - INFO - Parameters: ../DriverNetPy/DriverNet.py -n sampleInfluenceGraph.txt -m samplePatientMutationMatrix.txt -e samplePatientOutlierMatrix.txt
2019-05-15 16:45:13,799 - preprocess - INFO - Reading gene-gene interacions done.
2019-05-15 16:45:15,513 - preprocess - INFO - Reading sample-gene expression outliers done.
2019-05-15 16:45:16,410 - preprocess - INFO - Reading sample-gene mutation done.
2019-05-15 16:45:16,455 - preprocess - INFO - Building bipartite graph done.
2019-05-15 16:45:17,703 - __main__ - INFO - Driver finding done. Start significance test now.
2019-05-15 16:45:17,703 - permutation - INFO - Start permutation, using 8 threads.
2019-05-15 16:50:32,328 - permutation - INFO - Permutation done.
2019-05-15 16:50:37,932 - __main__ - INFO - Significance test done.
2019-05-15 16:50:37,937 - __main__ - INFO - Writing output done. Please check DriverNet_res.txt.
2019-05-15 16:50:37,938 - __main__ - INFO - All done. Cheers.

Compare the result of DriverNetPy with DriverNet.

python ../script/compare.py DriverNet_res.txt DriverNetPy_res.txt

Notes

The result of DriverNetPy shows discordance with that of DriverNet

This may due to:

the reasons aforementioned in the Motivation part.
when ranking candidate drivers, in case of a tie, randomly pick one.

Network permutation scheme

The network permutation scheme in DriverNetPy considers the degree of each node in the network. The degree of each node before and after the permutation are roughly the same.

The version of DriverNet I referred to was 1.0.0

Well, the code of DriverNet_1.0.0 (oldest) and DriverNet 1.24.0 (latest, checked in 20190515) shows no differences.

$ wget https://www.bioconductor.org/packages/2.12/bioc/src/contrib/DriverNet_1.0.0.tar.gz
$ tar zxf DriverNet_1.0.0.tar.gz; mv DriverNet DriverNet_1.0.0

$ wget https://www.bioconductor.org/packages/3.9/bioc/src/contrib/DriverNet_1.24.0.tar.gz
$ tar zxf DriverNet_1.24.0.tar.gz; mv DriverNet DriverNet_1.24.0

$ cd DriverNet_1.24.0/R
$ for sample in `ls *`; do diff $sample ../../DriverNet_1.0.0/R/${sample}; done

DriverNetPy was designed based on the sample data of DriverNet

I would change the forms of input data, e.g. from raw expression data and raw mutation data.