Network motifs

Posted on Tuesday 5 November 2002 to Frontiers

An international team of scientists said Thursday they have used a mathematical algorithm to detect recurring patterns in the networks making up everything from food webs to the Internet to gene regulation in cells.


By uncovering these crucial building blocks of networks, researchers have taken an important step toward unraveling the bewildering complexity of these systems, which they term "motifs."

"Understanding a motif's function may open the way to new ways of dealing with diseases for which there is no cure at present, diseases that are complex network-level malfunctions," researcher Uri Alon, a physicist at the Weizmann Institute of Science, told United Press International.

The research is based on the assumption that information and energy tend to flow in distinct networks. Certain plants and animals are eaten by specific lifeforms in food webs, for instance.

"21st-century sciences are obsessed with networks. The big problem is how to break down these complex networks into parts we can understand," Alon said from Belgium.

For instance, although scientists have mapped out where every human gene is, they do not yet fully understand how these thousands of genes interact, said physicist Albert-Laszlo Barabasi of the University of Notre Dame in Indiana. "In order to cure some of the major diseases such as cancer or depression, where several genes are functioning simultaneously, you need to understand the networks of the cells," he explained.

Using the algorithm, Alon and colleagues have developed a new experimental technique that maps out the wiring diagrams of these networks.

"We start with a network -- a list of elements and their connections," he said. "We then count how many times different patterns appear in this network. To understand which of the many patterns that occur are significant and potentially important, we compare the network to a large set of randomized networks.

These are networks ... made of the same elements but rewired so that the connections are scrambled. In each of the randomized networks we again count the number of appearances of the different patterns."

After a while, the computer program reveals some patterns occur much more often than they would at random. "These are likely to be patterns that are 'designed in' or 'highly selected for' by evolution," Alon said. [link]
This stuff makes me think of Stuart Kauffman's work on random boolean networks. He examined a wide range of random networks and discovered that only the ones that had an optimum amount of interconnectivity demonstrated interesting behaviour. Networks that weren't very well connected tended to freeze up and ones that were too well connected behaved chaotically.

This current work takes the opposite "bottom-up" approach. Rather than studying the behaviour of completely random networks, this group studied real networks found in nature such as gene networks, neural networks, food-chains etc. (as well as others like the Internet and electronic circuit diagrams). They scanned through the connections of each of these networks looking for the patterns that occur much more frequently than they would in a network that had been connected randomly. Only those patterns that occurred considerably more often in these living non-random networks were noted and by this decomposition technique were able to deduce the most important functional building blocks (design "motifs") that go in to making these networks tick.

But still, just like Kauffman's models, these patterns tend to be few in number and only connect a relatively few nodes at any one time and yet remarkably they manage to perform important and quite complex regulatory tasks. For a look at the actual research being described here, check out Uri Alon's website.