What IS StarNet?

StarNet is a visual data mining front end for exploring correlation networks constructed from mouse microarray data. This software serves two main functions: (1) it readily provides new hypotheses via the standard “guilt by association model,” where genes that participate in the same pathways frequently have similar expression profiles, and (2) it provides a starting place for reconstructing biological networks using modeling approaches such as dynamic Bayesian networks, by providing lists of candidate genes to use in such approaches. StarNet asks the user for a gene of interest, some parameters to specify how the network will be constructed, and some parameters for drawing preferences. StarNet returns two correlation networks local to the gene of interest: one that is constructed from correlations derived from 2145 microarray samples from a variety of tissues and experimental conditions (the ‘full cohort’), and one constructed from a subset consisting of 239 of those samples, which are derived from cardiac tissues or early developmental stages (the ‘cardiac cohort’). StarNet offers several useful features, including: network gene lists linked to Entrez gene, with flagging of genes found in both networks; edge (correlation) lists with 95% and 99% confidence intervals; optional highlighting and listing of nodes with specified GO ontology keyword matching (default=’transcription’); lists GO terms (and associated genes) that are enriched in the network compared with the entire array; and finally, to easily compare the correlation network to current knowledge, StarNet draws networks of known interactions (from Entrez’s Gene RIFs) involving genes in the drawn correlation networks.

How to cite

Jupiter, D.C. and VanBuren, V. A Visual Data Mining Tool that Faciltates Reconstruction of Transcription Regulatory Networks. PLoS ONE 3(3): e1717 doi:10.1371/journal.pone.0001717 [PDF] [Link to StarNet]

Getting started

The easiest way to see what StarNet does is to enter the symbol for your favorite gene and click submit. This will draw networks using our default parameters. If you don’t know the official symbol or Entrez ID for a gene, you can use the Mouse gene ID lookup tool on the StarNet front page to search by keyword or by partial symbols.

Data sets

Data from 2145 samples on the Affymetrix GeneChip Mouse Genome 430 2.0 array (GEO ID: GPL1261) were used in creating the correlation table we are calling the 'full cohort'. A subset of 239 arrays from the 'full cohort' was used in building the correlation table for the 'cardiac cohort'. There are 16297 reliable features on these arrays (see Data processing).

Data processing

We normalized and scaled the microarray data, and calculated correlation coefficients for all gene pairs (~130 million coefficients). We then partitioned the distribution of coefficients to get more tractable and meaningful subsets. Starting with the whole distribution, we first took the top 20000 (20K) coefficients from each tail of the distribution. We made similar distributions with the top 40K and top 100K coefficients. As there are many genes that are highly 'connected' in these tails, the 100K tails only contain ~3500 genes out of the total 16297. To get complete coverage of features on the array, we created a 'Genecentric' distribution, which contains the top 10 positive and top 10 negative correlations for every feature. We used the same 'top 10 positive/negative' approach to construct two specialty distributions: one where both genes have GO annotation matching 'transcription', and one where each gene matches either 'transcription' or 'signal'.

StarNet draws sub-graphs of larger correlation networks starting with your gene of interest in the center of the graph. Networks are drawn in concentric ‘levels’, where the first level consists of genes that are directly connected to the gene of interest. The second level consists of genes directly connected to genes in the first level, and so on.

StarNet can draw the following types of networks:

Levels - this draws every connection in the specified distribution for N levels, starting from your gene
Levels with Internal Edges - same as above, but connections within a level or back to a lower level are allowed
Weight - the user supplies a cutoff; the product of the coefficients in the path from your selected gene to any other gene must be higher than this cutoff; up to N levels are drawn, where N is user specified
Highest - draws the top n connections for your specified gene, then does the same for the next level, and so on; up to N levels are drawn
Highest with Internal Edges - same as above, but connection within a level or back to a lower level are allowed

Clicking on the resulting graphs will spawn a new page for that cohort, where the nodes are linked to NCBI's gene description. Below the graphs there is supporting information and analysis, including a gene list, an edge list with confidence intervals, a list of the genes in the network that match the user-supplied GO search terms (default ='transcription’), nodes with matching GO terms are also highlighted in red on the graph), GO terms enriched in the network compared with the whole array platform, and small networks of known interactions for genes in the correlation networks.

Complete documentation can be found in the User manual and the White paper

License information regarding the color palattes used in StarNet:

Apache-Style Software License for ColorBrewer software and ColorBrewer Color Schemes, Version 1.1
Copyright (c) 2002 Cynthia Brewer, Mark Harrower, and The Pennsylvania State University. All rights reserved.
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
1. Redistributions as source code must retain the above copyright notice, this list of conditions and the following disclaimer.
2. The end-user documentation included with the redistribution, if any, must include the following acknowledgment: This product includes color specifications and designs developed by Cynthia Brewer (http://colorbrewer.org/). Alternately, this acknowledgment may appear in the software itself, if and wherever such third-party acknowledgments normally appear.
3. The name "ColorBrewer" must not be used to endorse or promote products derived from this software without prior written permission. For written permission, please contact Cynthia Brewer at cbrewer@psu.edu.
4. Products derived from this software may not be called "ColorBrewer", nor may "ColorBrewer" appear in their name, without prior written permission of Cynthia Brewer.
THIS SOFTWARE IS PROVIDED "AS IS" AND ANY EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL CYNTHIA BREWER, MARK HARROWER, OR THE PENNSYLVANIA STATE UNIVERSITY BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.