As someone who has spent a lot of time staring at big hairball biological networks, I've always been frustrated by them. I've been looking for a better solution for some time now. I'm not sure this is it though.
My main beef is that I feel strongly that a visualization succeeds when you don't have to explain how it works. Just looking at the main page, without reading any of the slide deck, I had a hard time understanding what was going on. In fact, for the first simple example in the slide deck, I found the original network to be easier to grok than the new version. Things certainly appear more organized than when in a big hairball, but I still have trouble extracting any meaningful information.
As you probably know, Martin Krzywinski is also the author of the Circos software. Circos has been in development for much longer and addressed different kinds of issues.
This seems like a great method for ad hoc visualizations of multi-dimensional and graph-structured data. I could see building a tool that allows a user to pick a few dimensions and toggle between different orderings.
That said, I wish the authors had not used/brought up the notion of "visual analytics". Visualizations are great exploratory and illustrative tools, but I believe they leave much to be desired when it comes to doing the actual work of analysis. When looking for explanations --- particularly in large datasets --- is it not better to define some objective criteria which one can measure, rather than relying on easily tricked visual cues?
You're right that it is a problem, but defining objective criteria before knowing exactly what it is that you're looking for is potentially just as bad. This is part of the reason folks like Tukey defended statistical graphics early on: exploratory data analysis is tremendously important, and it's rare that you would know exactly what you're looking for.
Also, this problem of making sure what you're seeing is actually in the data is one that researchers are actively working on! There's a paper due to come out at a top visualization conference on the relation between formal hypothesis testing and visualizations which goes to the heart of your "easily tricked visual cues" comment. It's not currently available (and I'm not a co-author), but if you're interested, be on the lookout for "Graphical Inference for InfoVis"
Often a diagram like this can help spot problem areas or patterns that otherwise could go unnoticed. I would say a visualization like this would/could be the first step in analysis of a complex problem, followed up by a detailed analyisis of objective criteria...
Yep. I totally agree. I guess I've always thought of exploratory work as being a distinct part of analysis projects. That's why I tried to say that I think this would work as a great exploratory tool.
My main beef is that I feel strongly that a visualization succeeds when you don't have to explain how it works. Just looking at the main page, without reading any of the slide deck, I had a hard time understanding what was going on. In fact, for the first simple example in the slide deck, I found the original network to be easier to grok than the new version. Things certainly appear more organized than when in a big hairball, but I still have trouble extracting any meaningful information.