Go through the tutorials available on the site. Read the publication for use examples.
Please contact email@example.com
Please send information on exactly what you were doing, your browser (and version), and operating system, include a screenshot if possible to firstname.lastname@example.org
The visualization prioritizes which connections to display. You can adjust the network filter slider to show lower confidence connections.
You can adjust the Network Filter slider to increase the number of connections shown. This will show lower confidence relationships. If the relationships to your node of interest are still very low, you may want to try your query using another network.
One would expect that the N (default N=15) interactions with the highest weight would be presented, but this is not the case. Genes are normalized against their overall connections throughout the network for choosing which to display. Hyper-connected genes are thus down-weighted which reduces the chances of hairball graphs.
Yes. Please contact email@example.com. This is not available by default to keep the site easy to use and because these networks are very large files.
Try to choose the network that is most appropriate for the area of inquiry. If you are not sure, you can use the human global immune network or you can try several of the specific networks and explore the results.
All connections are inferred from the data compendium. The known connections are used to train the network, but if any “known” connections are not supported by the data, they will have extremely low weights and will not be displayed.
If you click on the edge of the network, you can see the evidence used and relative weight for making each edge prediction.
The Bayesian integration methods utilized handle these issues automatically by weighing all underlying data according to how useful they are for making predictions in the training set. All data of any type that improves prediction is utilized and weighed accordingly. In the example shown in Table S2 in the paper, for example, many types of data are highly weighted in making inferences. This is what is typically found. We have found no evidence that any type of data drowns out any other type of data in making predictions.
The ImmuNet networks are available for download as tab-delimited, gzipped files here. These networks include all edges with a posterior probability above the prior(0.005). They are presented in the following format:
[Entrez GeneID 1] [Entrez GeneID 2] [Posterior Probability]
Once the files are uncompressed they may be opened as tab delimited files with Excel or a text editor. Due to the large size of these network files we recommend using the ImmuNet interface for any network visualization.
Gorenshteyn D, Zaslavsky E, Fribourg M, Park C, Wong AK, Tadych A, Hartmann BM, Albrecht RA, Garcia-Sastre A, Kleinstein SH, Troyanskaya OG, Sealfon SC. Investigating immunological pathways and diseases with a comprehensive compendium of human data. Immunity. 2015