ims-biclustering

Biclustering algorithms for processing imaging mass-spectrometry data.

Installation

To compute eigenvalues, ims-bicluster uses ARPACK; the ARPACK++ headers are included. The following libraries have to be installed: libblas libsuperlu libarpack libgfortran libgfortranbegin libnsl To install them on a Debian-like system, run sudo apt-get install libarpack2-dev libsuperlu3-dev gfortran libatlas libatlas-dev libblas-dev libboost-system-dev libboost-date-time-dev libboost-filesystem-dev

To read Matlab input files, ims-bicluster uses the matio library. It can be obtained from http://matio.sourceforge.net/

Reports generation uses LaTeX and python with numpy and matplotlib packages: sudo apt-get install latex texlive dvipng libfreetype6-dev libpng-dev sudo pip install matplotlib numpy scikit-learn statsmodels

Usage: bash scripts

There are two bash scripts for the IMS dataset processing cycle:

./gen-all.sh fname

Expects to find input file in data/fname as a Matlab .mat file. Runs ims-bicluster, then runs gen-report to generate the report pdf.

./gen-report.sh fname

Expects to find input file in data/fname as a Matlab .mat file together with the following files produced by ims-bicluster: fname.mat.coords.csv -- coordinates of spectra on the image fname.mat.val.csv -- eigenvalues fname.mat.vec.csv -- eigenvectors Runs ims-bicluster, then runs gen-report to generate the report pdf.

Usage: ims-bicluster

 ./bin/ims-bicluster --mat --eigens=n --input=ims-dataset.mat

Computes n largest eigenvalues for the matrix encoded as a Matlab .mat file with the following structure:

• x_printed -- vector of X coordinates of all pixels in the dataset (len=num_pixels)

• y_printed -- vector of Y coordinates of all pixels in the dataset (len=num_pixels)

• spectra -- matrix of intensities for each pixel (shape = num_pixels x len_spectrum)

  ./bin/ims-bicluster --mat2 --eigens=n --input=ims-dataset.mat

Computes n largest eigenvalues for the matrix encoded as a Matlab .mat file with the following structure:

• x -- vector of X coordinates of all pixels in the dataset (len=num_pixels)
• y -- vector of Y coordinates of all pixels in the dataset (len=num_pixels)
• SP -- matrix of intensities for each pixel (shape = num_pixels x len_spectrum)

Usage: visualize.py

python python/visualize.py -e data/fname.val.csv -v data/fname.vec.csv -c data/fname.coords.csv -N num_clusters -M num_eigenvalues

Performs clustering for all numbers of clusters from 2 to num_clusters and num_eigenvalues largest eigenvalues based on ims-bicluster processing results; generates a pdf report. The LaTex code of the last processed report is stored at reports/latex/tmp.tex.