A picture is worth a thousands words, or so the saying goes. What it really should be is a good picture is worth a thousand words. As a scientist, I use visualizations in the form of diagrams and flow charts to illustrate difficult concepts and accompany technical descriptions.
Why Hive plots?
Visualizing large networks is tricky; conventional graph layout such as force directed layouts are inconsistent, often resemble “hair balls” and can even suggest patterns which aren’t actually there. Comes in Martin Krzywinski (Genome Science Center, Vancouver, BC) and develops hive plots, a coherent network layout algorithm that places nodes using predefined rules. Hive plots facilitate the visualization of structural patterns in a network and the comparison of networks. The figure below illustrates how hive plots facilitate the discovery of social patterns in this fake social network between boys, girls and aliens.
On the left we see the usually force directed layout of the social network with people colored by “gender” and interactions between people are colored according to their friendship type “friends” or “enemies”. On the right, the hive plot organizes the people onto axes by gender and along the axes by the number of interactions they have in the network: the more interactions the further they are from the center.
There are three patterns that are easily discernible in the hive plot but not in the force directed layout (labeled in the figure): girls and boys don’t interact (pattern 1), boys are all enemies (pattern 2) and the person with the most interactions is the alien named Zans (pattern 3). How easily can you find these patterns in the force-directed layout version? Now imagine if you are looking at your own Facebook social network with hundreds, maybe even thousands, of connections: these kinds of social structural patterns will be impossible to find in the massive “hairball” produced by a force directed layout!
Why Hive Panel Explorer?
Why make one hive plot when you can make many? Why create static viz when you can interact with the data?
A hive panel is exactly what it sounds like: a layout of multiple hive plots, all representing your network in a different way. These multiple projections come in handy when you have “big data” : you can pick multiple properties of nodes (degree, clustering coefficient, gender in social networks, protein family in PPI networks etc.) to organize the nodes and layout the network differently on each plot. In this way, you can compare the positions and edges between nodes on different hive plots.