Biclustering fMRI time series: a comparative study

This repository aims to provide support to the associated paper. It acts as an repository to archive:

• The scripts that generate the synthetic datasets
• The datasets used in the paper
• Scripts to generate the biclusters
• Functions to read and analyse results
• Some generated biclusters to act as example

Repository structure

• SimTB Generators: Related with the process of generating synthetic datasets
• Data Collections: Data used for the analysis
• Biclustering Generators: Scripts for generating biclusters from the data
• Biclustering Analysis: Functions to compute the analysis metrics

Data collections

The paper have three data colections.

• First data collection: A single synthetic subject
• Second data collection: Twenty synthetic subjects
• Third data collection: Twenty real subjects

Algorithms

Biclustering algorithms

• ISA
• XMotifs
• FABIA
• Spectral
• Bimax
• CCC
• BicPAM

Clustering algorithms

• K-Means
• Spectral
• Ward's

Workflow

Software requirements

During the development of this paper some software was needed depending on the part of the analysis:

• MATLAB will be needed to run the SimTB fMRI data simulator, to generate the synthetic datasets.
• R will be needed for running most of the biclustering algorithms (FABIA, ISA, Bimax, XMotifs, Spectral).
  • For FABIA, you need the fabia Package.
  • For ISA, you need the isa2 Package.
  • For the remaining algorithms, you need the biclust Package.
• Python will be used for running the clustering algorithms (Spectral, K-means, Ward's hierarchical method) and analysing the biclustering solutions.
  • For the clustering algorithms, we used the implementations provided by scikit-learn.
  • Other traditional python libraries are used during the scripts.
• BicPAM is implemented in the BicPAMS tool.
  • Both an Desktop GUI and a Java API are provided.
• CCC is implemented as part of the BiGGEsTS software (GUI implementation).