Skip to content

realmarcin/MAK

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Massive Associative K-biclustering -- MAK

Installation

Yeast column bicluster single CPU example (precomputed nulls)

MAK HPC bicluster discovery pipeline

Installation

Requirements:

  1. Java (tested with version 11.0.16)
  2. R, compiled with --enable-R-shlib (tested with 4.2.1)
  3. Four R packages including rJava which installs JRI, the Java-to-R interface

The installation has been tested in the bash shell on Linux and Mac OS.

Installation steps:

  • Use the MAK.jar file found in this repo
MAK_build/MAK.jar

or build it from scratch

cd MAK_build/
source antbuild_git.sh
  • Install R package dependencies with these R commands on the R command line:
> install.packages("fields")
> install.packages("amap")
> install.packages("irr")
> install.packages("rJava")
  • Once rJava is installed you can find the location of the JRI.jar file using this R command on the R command line:
> system.file("jri",package="rJava")
  • Add the JRI and MAK JAR paths to your system CLASSPATH variable, e.g in linux bash:
export CLASSPATH=$CLASSPATH:path/to/JRI/JRI.jar:path/to/MAK.jar
  • Check/set your R_HOME variable, e.g. in linux bash: To find you R home dir use this R command on the R command line:
> R.home()

And use this path to set the $R_HOME environment variable:

export R_HOME=path/to/R
  • Add JRI shared library files and R shared library files to the LD_LIBRARY PATH variable:
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:path/to/JRI/:$R_HOME/lib

Running a single bicluster search using precomputed null distributions

  1. Download the MAK yeast example tar ball: https://github.com/realmarcin/MAK_results/raw/master/results/yeast/input/example/MAK_example.tar.gz
  2. Place this tar ball in the top level MAK repo directory and extract it:
tar zxf MAK_example.tar.gz
  1. Run a single MAK trajectory (about 30 min. on a MacBook Pro 2.4Ghz Intel Core i9):
cd MAK_example
source run.sh

If you get a java library path error, you can specify this path in the java command:

java -Djava.library.path=path/to/JRI/ -Xmx2G DataMining.RunMiner -param MAK_parameters.txt -debug 0
  1. You can perform searches for any starting point in this dataset by changing the value of the INIT_BLOCKS field in the MAK_parameters.txt parameter input file. The indices are 1-offset and based on the row and column labels in the yeast_cmonkey.txt input file. The format of the value is:
{a,b,c/x,y,z}

where a,b,c,x,y,z are integers in the range of the dimensions of the input matrix.

Running the MAK HPC bicluster discovery pipeline

  1. Input data:
  • Simple TSV file(s) with row and column labels (with shared identifier axes if multiple data layers)
  1. Edit fields in MAKflow input parameter file:

E.g:

  • Input TSV data path(s) (up to four)
  • Bicluster coherence criterion (column = MSEC_KENDALLC_FEM)
  • null size boundaries (y=(2,200), x=(2,100))
  • starting point size range (y=(10,100), x=(10,50))
  • HCL starting point metric and linkage (Pearson correlation and Complete)
  • move mode sequence (BS)
  • HPC parameters
  • number of CPUs (600)
  • maximum wall time (48 hours)
  1. Run:

java DataMining.MAKflow ...

Syntax: -parameters = input parameter file -server = HPC cluster -account = HPC user account -qos = HPC queue

  1. Output:
  • A TSV file of nonredundant, highest scoring biclusters in the input data
  • A set of directories with all intermediate data and results
  • Includes all executed commands and optional logging

About

Massive Associative K-biclustering

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published