DREAM network inference challenge

The DREAM initiative organizes an annual reverse engineering "competition" (we prefer to see it as a community experiment) called the DREAM challenges. We use GNW to provide the so-called DREAM network inference challenge.

The goal of this challenge is to predict network connectivities from real (in vivo) and simulated (in silico) gene expression datasets. The true structure of the gene networks is revealed only after the submission of the predictions. In addition to inference of network structure, we also encourage prediction of gene expression measurements for specific conditions that we withhold from the training data.

Below, we describe the three editions that we provided so far:

• DREAM5 network inference challenge (2010)
• DREAM4 in silico challenge (2009)
• DREAM3 in silico challenge (2008)

DREAM5 network inference challenge (2010)

The DREAM5 network inference challenge is part of the DREAM5 conference, to be held at Columbia University on November 16-20.

This year's challenge is literally a challenge: the goal is to infer the genome-scale transcriptional regulatory networks of three microorganisms. As input data we provide a unique compendium of over 1500 microarray chips. In addition, the challenge also contains an in silico network, which will allow thorough performance evaluation.

The description and data of the challenge are available on the DREAM website:

• DREAM5 network inference challenge

For convenience, the following archive contains the expression data of the challenge in two alternative formats (the transpose of the data matrix, and with multiple replicates of the same experiment averaged):

• Alternative data formats.zip

DREAM4 in silico challenge (2009)

The DREAM4 in silico challenge was associated with the DREAM4 conference, held at the Broad Institute of MIT and Harvard in December 2009. There were 33 participating teams, which submitted a total of 300 network predictions for the various benchmark networks of the challenge.

As in the DREAM3 challenges, participants were asked to infer network structures from the given in silico gene expression datasets. In addition, we encouraged participants to use their inferred models to predict the response of the networks to novel perturbations in an optional "bonus round" of the challenge. A detailed description and the results of the challenge are available on the DREAM website:

• DREAM4 challenge description
• DREAM4 results (rankings, gold standards, evaluation scripts)

Videos of DREAM4 presentations are available in the NYAS eBriefing:

• DREAM4 in silico challenge introduction (Daniel Marbach)
• DREAM4 challenge results (Robert Prill)
• Presentations of the three best performers (Robert Küffner et al., Nicola Soranzo et al., Alex Greenfield et al.)

The following archive contains all challenges (datasets + gold standards) as available on the DREAM website, plus additional information (additional datasets, the datasets without noise, details on the applied perturbations, the signed goldstandards, network graphs in EPS format, etc). See the pdf file for details:

• DREAM4 in silico challenge.pdf
• DREAM4 in silico challenge.zip

DREAM3 in silico challenge (2008)

The DREAM3 in silico challenge was held with the DREAM3 conference at the Broad Institute of MIT and Harvard in October 2008. We provided three sub-challenges with networks of sizes 10, 50, and 100, respectively. 29 teams participated in the in silico challenge. They submitted 390 network predictions in total. For details, see the DREAM website:

• DREAM3 challenge description
• DREAM3 results (rankings, gold standards, evaluation scripts)

See also the videos of the DREAM3 presentations (also available as NYAS eBriefing, you may need RealPlayer):

• Daniel Marbach. DREAM3 in silico challenge introduction
• Gustavo Stolovitzky. DREAM3 challenge results
• Kevin Yip. DREAM3 in silico challenge best performer

The following archive contains all challenges (datasets + gold standards) as available on the DREAM website, plus additional information (network graphs in EPS format, datasets without noise, signed gold standards, etc). See the README file for details:

• DREAM3 in silico challenge.zip