### Command prompt is indicated as $

1) Extract code and data. 

$ tar -jxf fpclass.tbz

A directory structure will be created under a root directory called 'fpclass'.


2) Compile the code

$ cd fpclass/modules
$ make
$ cd ../moduleIntegration
$ make
$ cd ../

3) Create input data.

Create a tab-delimited text file with IDs of protein pairs - interaction predictions will be made for these pairs. 
Each line in the input file should have the format: <UniProt ID1>	<UniProt ID2>	<class label>
e.g., P10809	Q15303	1
The protein IDs in each row should be ordered alphabetically (e.g., P10809	Q15303 rather than Q15303	P10809). This can be done with the perl script 'orderProteinPairs.pl', found in the /fpclass directory.
The class label was necessary for software development and testing - it can be set to any integer value, and does not currently affect the output or execution of fpclass.
An example input file, 'exampleInput.txt', is present in the /fpclass directory.

4) Run FpClass

perl runFpClass.pl <file with protein pairs> <prefix for output files>
For example,
$ perl runFpClass.pl exampleInput.txt example

The terminal output will look as follows:
$ Running FpClass on exampleInput.txt, output files will start with 'example'
$   paralogs

Runtime was 15 minutes for 100 protein pairs, using an 8 core Intel Xeon E5345 2.33 GHz CPU. About 2 GB of memory were required.


5) Output from FpClass

Two main output files are produced:
	
	/fpclass/example.finalResults.probabilities.txt
	- for each protein pair, shows the overall probability of interaction, and probabilities based on individual evidence types
	
	/fpclass/example.finalResults.scaledProbabilities.txt
	- same as the previous file except probabilities from individual evidence types are scaled to highlight their relative importance. The scaled probability for each evidence type i, scaled_probability_i, was calculated as the overall probability of interaction multiplied by the probability from evidence type i, normalized by the sum of probabilities from all evidence types:	
	
		scaled_probability_i = overall_probability * (probability_i / sum(probability_j) for j = 1..evidence types)
	
Note that probabilities estimated via coexpression and overall probabilities may differ from those reported in the supplementary materials. This is due to older gene chip annotations used in the calculation of supplementary material results.