SUPPLEMENTARY MATERIAL FOR THE MANUSCRIPT:

"RGIFE: a ranked guided iterative feature elimination heuristic for biomarkers identification"

This folder contains the supplementary material that supports the manuscript "RGIFE: a ranked guided iterative feature elimination heuristic for biomarkers identification".

We provide the material used for the biological validation of our approach using a prostate cancer dataset (Singh et al. 2002) as case study.

SIGNATURES:

We present all the signatures used in Section 3.4 of the manuscript to assess their relevance in a biological/disease context. 
Our 3 policies (RGIFE-Union, RGIFE-Min and RGIFE-Max) were compared with 4 well known feature selection algorithms (CFS, SVM-RFE, Relief, Chi-Square). 
The genes that define the signatures can be found in the signatures_case_study.tar.gz archive. 
Here is a summary of the files:

Method        Number of genes    Filename

RGIFE-Union   21                 RGIFE-Union.txt
RGIFE-Max     12                 RGIFE-Max.txt
RGIFE-Min      7                 RGIFE-Min.txt
CFS           88                 CFS.txt
SVMRFE        21                 SVMRFE.txt
Relief        21                 Relief.txt
Chi-Square    21                 Chi-Square.txt


SIGNATURE INDUCED NETWORK:

A part of the biological confirmation of our signatures involved the analysis of the genes in a network context. 
We generated a signature induced network from a PPI network by aggregating all the shortest paths between all the genes in the signature. 
The genes used for the network inference were inferred by RGIFE-Union (the best performing RGIFE policy). 
In the file signature_induced_network.pdf we provide a visualisation of the signature induced network. 
The network is available in edge-list format from signature_induced_network_edges.txt.