FSWeight.Zip
1 March 2007
============

The source codes for FSWeight.Zip along with the VC project files are 
provided here. The program runs in windows 2000/XP commandline. 
There is a sample batch file and some sample input files. 


INPUT
The software will take the following command:
relate scheme_file annotation_file interaction_file

scheme_file
 - Function scheme file (e.g. funcat-2.0_scheme.txt)
 
annotation_file
 - Function annotation file (e.g. funcat-2.0_data_14032005.txt)

interaction_file
 - Protein interaction file (e.g. grid_1804205.txt)

Refer to the files for the respective formats. 
All files are simple delimited flat text files.


OUTPUT
The software will output 2 files: 

predictions.txt
 - contains predictions made in this format (tab-delimited):
	ProteinName   FunctionID   Weight
  
results.txt
 - contains the data for the precision vs recall graph (tab-delimited):
	Score  Recall  Precision      


OPERATION
By default the software will perform leave-one out cross validation 
using FS-Weighted Averaging.

Some simple comments are included in main.cpp on how to:

1) Perform X-fold cross validation

2) Use Neighbour counting, Chi-Square and Functional Flow. 

Functional Flow  is implemented based on the description in the 
manuscript (Nabieva,E. et al. (2005) Whole-proteome prediction of
protein function via graph-theoretic analysis of interaction maps.
Bioinformatics, 21 (Suppl. 1), i302¨Ci310). We received comments
that the authors made some mistake in the description, but was not 
given any details. Hence it would be best to request the software
from the authors.

3) Output the distance matrix for the clustering in PRODISTIN. 
(Brun,C. et al. (2003) Functional classification of proteins for
the prediction of cellular function from a protein¨Cprotein 
interaction network. Genome Biol., 5, R6.). This matrix will be 
output to the file distmatrix.txt. This can be used as an input 
for the BIONJ clustering program which is also included here. 
BIONJ will output a tree as a text file. Name this file as tree.txt, 
then use the function PrintProdistin() to make predictions from 
the tree.

MRF

We did not implement the MRF method (Deng,M. et al. (2003) Prediction 
of protein function using protein¨Cprotein interaction data. 
J. Comp. Biol., 10, 947¨C960). Instead we run our software on their 
data and compare with their results. I have included the formatted 
functional scheme, annotation and interaction files, as well as 
precision vs recall results of the MRF method here in the folder 
"MRD data". Original data files used to be available at:
http://www-hto.usc.edu/ˇ«msms/ProteinFunction, 
but do not seem to be available anymore.


Credits:

The FSWeight protein function prediction program was implemented
by Hon Nian CHUA. If you use this software, please cite the following
paper:

Hon Nian Chua, Wing-Kin Sung, Limsoon Wong. 
"Exploiting Indirect Neighbours and Topological Weight to 
Predict Protein Function from Protein-Protein Interactions". 
Bioinformatics, 22:1623--1630, 2006.