[Cytoscape-discuss] RCytoscape: speed improvement sending large networks to Cytoscape

[Paul Shannon]

A message I just sent out to cytoscape-discuss.

Cheers!

- Paul

There is at least a small audience of RCytoscape users among those who follow this discussion group. I have heard concerns about slow transfer speeds between R and RCytoscape for large graphs. I hope this report (and bug fix) will be good news to them, and maybe interesting to others.

RCytoscape's "displayGraph" function sends a graph constructed in R to Cytoscape.

Due to some rather dumb coding choices I made a year ago, large graphs (e.g., 5000 nodes, 45k edges) took extraordinarily long times to transmit from R . I usually work with smaller networks, so I had not paid enough attention to this. I received a few reports in the last couple of weeks of these long transmit times, found the error of my ways, and now report that they are much improved. RCytoscape 1.6.4 will be available tomorrow (Tuesday) at

http://bioconductor.org/packages/2.10/bioc/html/RCytoscape.html

In addition, the next version of RCytoscape, 1.7.9, available soon in bioc-devel, will have a new method, displayGraph.predictedTime.

It will use this formula, alculated by R's linear model function "lm", applied to timing data I just collected:

transferTime = (0.002 * node.count) + (0.013 * edge.count) + (0.002 * noa.count) + (0.002 * eda.count)

where:

node.count = length (nodes (graph))
edge.count = length (edgeNames (graph))
noa.count = length (noa.names (graph)) * node.count
eda.count = length (eda.names (graph)) * edge.count

Two examples, one small, one big:

A 500 node, 500 edge graph transmits from R to Cytoscape in about 15 seconds.
A 5k node, 45k edge graph with 3 node attributes, and 3 edge attributes, will transmit to Cytoscape in about 10 minutes.

The behavior of the three elements of software involved here (Cytoscape, XML-RPC, R) is not quite linear. This formula underestimates the transmission times for small graphs, and gets the time about right for very large ones.

Cheers,

- Paul Shannon
Bioconductor Project
Fred Hutchison Cancer Research Center, Seattle






--
You received this message because you are subscribed to the Google Groups "cytoscape-discuss" group.
To post to this group, send email to cytoscap...@googlegroups.com.
To unsubscribe from this group, send email to cytoscape-disc...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/cytoscape-discuss?hl=en.