It is important to have a standard set of known interactions to compare different algorithms in a uniform way. We have focused on experimentally determined interactions as opposed to synthetically generated networks. Although synthetic networks are valuable for testing parameters and other features of network inference algorithms, there is already a large amount of work done in this area.
What is more difficult is finding a large set of experimentally determined regulatory interactions to test against. E. coli is currently the only species a with large number of freely available regulatory interactions (collected in RegulonDB).
Data
The most up-to-date microarray data can be found on the M3D website. However, below is the data from E_coli_v3_Build1, which was used for the paper. For most of the algorithms, we used the RMA normalized data. The workspace below requires matlab 7.0 or later.
Download the E_coli_v3_Build1 workspace
Download the E_coli_v3_Build1 workspace README file
We also used a list of transcription factors that we compiled from the known and putative transcription factors available on RegulonDB.
Download the transcription factors used in the paper
Validation
To score network inference algorithms, we used known transcriptional regulatory interactions available from RegulonDB.
Download the known regulatory network used in the paper
You can use the CLR output matrix for all 4345 genes to compare your results with CLR.
Download the CLR inferred regulatory network used in the paper
Also, here are two text files containing the CLR predicted connections from the PLoS paper at 60% precision and 80% precision.