Welcome to the Prophetic Granger Causality repo! This code allows you to recreate the results of experiments presented in the work by Carlin, et. al. [1]
If you use the GENIE3 portion of the code, please cite the corresponding method [2].
You will have to install R (http://www.r-project.org).
For any of the GENIE3 variants, you must also install the R package randomForest [1]. This installation can be done from R with this command (as root under Linux):
> install.packages("randomForest")
You will need python 2.7 for the prior heat kernel calculations, along with these packages:
- numpy
- scipy
- optparse
The complete dream8 winning submission can be recreated with:
make dream8
To use PGC, you will need to at least do:
make lasso.so
Within R, the data methods take a single data cube as input. This cube Z is a list of n by m
matrices, where each element of the list is a replicate of n time steps by m genes. The result that
is returned is a single m by m matrix of gene-gene interaction predictions.
Begin by executing the following:
source('dream8-granger-full.R')
source('prophetic-Genie3.R')
From here, if you are interested in PGC, use the following:
PGC_result=full.granger(Z)
To use Prophetic GENIE3, use the the following instead:
PG3_result=prophetic_GENIE3(Z)
See also the synapse for this project:
https://www.synapse.org/#!Synapse:syn2347433/wiki/62276
[1] Daniel E. Carlin, Evan O. Paull, Kiley Graim, Chris Wong, Adrian Bivol, Peter Ryabinin, Kyle Ellrott, Artem Sokolov, Joshua M. Stuart. "Prophetic Granger Causality to infer Gene Regulatory Networks."
[2] Van Anh HUYNH-THU, Alexandre IRRTHUM, Louis WEHENKEL, Pierre GEURTS. "Inferring regulatory networks from expression data using tree-based methods." PLoS ONE vol. 5(9): e12776
[3] Breiman L (2001) "Random forests." Machine Learning 45: 5-32. http://stat-www.berkeley.edu/users/breiman/RandomForests/