I'm fully aware such massive networks are not "uncommon", but I very often question their value or relevance of them in a paper or presentation.  It seems to me, far more often then not, most of what is included is wholly irrelevant to the discrete biological question(s) being asked and supposedly addressed with the data.

Just because one has software tools that can handle such a massive dataset does not mean that dataset is optimal for the biology being investigated, and certainly not necessary for the presentation of a biological interpretation.

In the absence of a clear rational for not pruning the network back right from the start, I question the relevance of trying to include all that data at once.

Since, as mentioned, nobody will be able to make any sense of it without massive simplification anyway, if there is a biologically based rational for simplifying the data right from the start, I'd want to think about that before plunging in. Presumably this network was generated to address some specific hypotheses, or at least some discreet functional biology question(s).

The short answer is that there isn't one. Large, complex networks will always look complex to the human eye no matter what layout you use. If you're trying to show something specific about the network, though, then a single visual property might help. Even a complex network can appear (slightly more) simple if each node is only one of two colors. 

Filtering helps, too. It may not even make sense to include every edge in the network unless your overall conclusion is "this network has almost 50 thousand edges."

That being said, you may want to try the AllegroLayout module.

http://apps.cytoscape.org/apps/allegrolayout

One method might be to sub-cluster nodes within the network based on some measurement you have relating to their function/annotation/transcriptional activity/etc. 

For example, take a look at this Cell paper: 

http://www.ncbi.nlm.nih.gov/pubmed/22939629

You might then have nice chunks to look at. Good luck

http://www.ncbi.nlm.nih.gov/pubmed/22939629

I guess I am just confused over what the point of this network is to begin with?  That is a truly ridiculously massive number of nodes and edges to try and show, let alone interpret, in any biologically meaningful presentation.  Not to mention it sounds like you have a huge number of redundant edges (more than 3 for every single node??)?  Is that true?

As already mentioned, without massive simplification, that network will never be displayable in any manner which any person will follow or make sense of.

How exactly did you arrive at this huge transcriptional network?  What is your hypothesis or biological question you want to illustrate with this network?  My guess would be you could throw out 90%+ of that data and focus on what is truly biologically meaningful given your initial experiment and its intent.

there are reference articles discussed about biological network with cytoscape, hopefully can help your research.

Hi,

     my recommendation is Circos using R or Java-script and apply filtration to give proper, collaborative experience to audience!

Regards

muddsair

HI, head/tail breaks would help you out; see the following paper

Jiang B. (2015a), Head/tail breaks for visualization of city structure and dynamics, Cities, 43, 69-77, Preprint: http://arxiv.org/ftp/arxiv/papers/1501/1501.03046.pdf

drop out the tail recursively, and you would be able to see the whole from the head parts.

When dealing with very large networks, clever thing to do is not to visualize as it is. If you do it will be a "hair ball" to start with and probably crash/freeze Cytoscape. You could however, use grouped node, visualization options and filters to highlight what is important.

When we faced the issue 5 years a go when we wanted to use all known protein interactions with Cytoscape, we built a plugin. It allowed to perform analysis on the entire network without visually showing the network and  draw selected parts when required. and to collapse sets of nodes to a single one.   The plugin works only with old Cytoscpe, but the open source code is available  http://irefindex.org/wiki/index.php?title=iRefScape

Yes, indeed. The pattern with the left panel looks like a hair ball, while the pattern with the right panel emerges. Note that the right pattern reflects the whole, which is invisible with the left panel. From the example, we can see how head/tail breaks enables us to visualize BIG data: http://en.wikipedia.org/wiki/Head/tail_Breaks

In your network edges ( 3 times) are quite less compare to the no. of nodes which indicates that the network is quite sparse and if you cleverly use some of the plugins ( MCODE and other clustering ones) you can see the meaningful layers/connections in your network. I use many in combination. Also it depends on your problem and by selecting the biologically important nodes, you can select the neighborhood in Cyto then create sub networks...like wise...

@Michael B Black: Infect its not too big network. 15 to 20 K nodes are quite common. Thats why networks to see the big picture from sky and then slowly reach the ground with cleverly clustering the related ones...Cyto handles it well...

I'm fully aware such massive networks are not "uncommon", but I very often question their value or relevance of them in a paper or presentation.  It seems to me, far more often then not, most of what is included is wholly irrelevant to the discrete biological question(s) being asked and supposedly addressed with the data.

Just because one has software tools that can handle such a massive dataset does not mean that dataset is optimal for the biology being investigated, and certainly not necessary for the presentation of a biological interpretation.

In the absence of a clear rational for not pruning the network back right from the start, I question the relevance of trying to include all that data at once.

Since, as mentioned, nobody will be able to make any sense of it without massive simplification anyway, if there is a biologically based rational for simplifying the data right from the start, I'd want to think about that before plunging in. Presumably this network was generated to address some specific hypotheses, or at least some discreet functional biology question(s).

Is the network fully connected, or are there subnetworks that are disconnected from each other?  If the latter, you might start by discarding small subnetworks with less than some threshold number of genes.  Also, you should consider finding hub genes and their neighbors within a few edges and extract those subnetworks and display them.  If you have other data that you can superimpose on the nodes, that might give some of them more relevance than others, that would be a way to find relevant subnetworks.  I find the igraph package in R an easy way to work with graphs, if you know R.  Because Cytoscape doesn't let you update a network with new data after laying it out, I find it annoying to work with.  I think of it as a last-step visualization tool.

Depende of your analysis. You can Use cluego in cytoscape and use a more stringent filter so you end up with the more relevant network.

Hi, You can use head/tail breaks to filter out those less connected nodes; see http://en.wikipedia.org/wiki/Head/tail_Breaks

@Bin Jiang - Is there any easy implementation of this algo. Is it integrated with any - Cytoscape or R based packages etc?

I usually use the ORGANIC LAYOUT to try an identify isolated subnetworks on my genome-scale reconstruction (and to show-off my reconstruction to people ofcourse...heh)

I would suggest you try to REMOVE CURRENCY METABOLITES such as ATP, NADP, CoA (acetyl-CoA), H2O, etc. Helps simplify the network a lot.

One easy way of doing it in the case of big networks is by sorting the metabolites based on the node degrees. These metabolites usually on the top of the list when you sort  the list by this attribute.

I havent yet tried sucha huge network on it but I also use the metexplore webtool (www.metexplore.fr)  to identify relevant subnetworks (I select my starting and ending nodes and the shortest, lightest path will be highlighted). I find it useful when trying to identify pathways based on marker metabolites identified in metabolomics experiments.

(Disclaimer: I work with the team so my opinion of metexplore might be biased) :-)

metexplore.fr

@ Sanjeev Kumar · 8.36 · 12.35 · BioCOS Life Sciences

@Bin Jiang - Is there any easy implementation of this algo. Is it integrated with any - Cytoscape or R based packages etc?

Usually you could do manually after you have made the head/tail breaks classification using the small software program: https://github.com/digmaa/HeadTailBreaks

Simply drop out the tail again and again until the graph structure is clear enough.

See examples in these papers

Ma D., Sandberg M., and Jiang B. (2005, accepted), Characterizing the heterogeneity of the OpenStreetMap data and community, ISPRS International Journal of Geo-Information, xx(x), xx-xx, Preprint: http://arxiv.org/abs/1503.06091

Jiang B., Duan Y., Lu F., Yang T. and Zhao J. (2014), Topological structure of urban street networks from the perspective of degree correlations, Environment and Planning B: Planning and Design, 41(5), 813-828, Preprint: http://arxiv.org/abs/1308.1533

I would define a couple of attributes for all the nodes. Then. I would color nodes by different attributes, and then visualize the network using Organic layout. This method gives you a quick way for big network visualization given your attribute of interest.

Since you are dealing with a transcriptional network, which has directed edges, you can use some of the automatic layout algorithms in BioTapestry (www.BioTapestry.org) to lay your network out. Using the alternating or the diagonal strategies will separate the control kernel from the pure targets in the network, and you can lay the targets nodes out so that nodes with identical inputs are adjacent to each other. The control structure will be revealed in the control kernel.

However, things will get sluggish in BioTapestry with 15842 nodes and 49812 edges. But BioFabric (www.BioFabric.org) will have no problems dealing with that size of network. In the case of this network, you will want to create a custom layout node attribute file to organize the nodes so that the control nodes (i.e. nodes with outputs) are placed in the top row, while targets are placed below. For an example of a GRN presented by BioFabric, see: http://vizbi.org/Posters/2014/Y06 Unlike the force directed approach, all sorts of network features are visible in this BioFabric global view.