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

3. 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

4. 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)

2. 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)

3. Run:

java DataMining.MAKflow ...

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

4. 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