A recently published paper in Nature Biotechnology describes a simple algorithm which can potentially cut through the background noise often found in larger networks.
Networks are huge, and they are everywhere now. Your Facebook profile is part of a large network of friends, likes, and interests. In biotechnology, for example, networks are used to make sense of interacting genes. What often happens though is that items which are close (either physically or connected through a third party) are mistaken for being related. Just because two people may know the same person, doesn’t mean that they know each other. These indirect relationships can create a feedback type loop which makes them appear stronger, often obscuring the more direct connections.
There are many ways to deal with these issues, however they are specific to their problem domain and often require a great deal of computational power.
Soheil Feizi et al. (2013) have developed an algorithm which located those stronger direct connections and filters out the indirect ones. They test their algorithm by applying it to three datasets; gene expression regulatory networks, protein structure predictions, and social network of co-authorship information.
What I particularly like about this paper and the authors is that they released their code and datasets. The code is a simple MATLAB module which can easily be used by other researchers and techies interested in this type of thing. The method they describe could also be applied to a wide variety of areas which makes it quite an exciting prospect (if you’re into that kind of thing!)
Many systems (including Artificial Intelligence research) is moving away from rule based methods, to one where relationships are built by analysing big data. Think Google Translate. It isn’t built from an endless list of rules about grammar and syntax, instead it uses a vast array of multi-language data (e.g. books printed in multiple languages) to build a network of related words and phrases. Amazon’s recommendation engine works in a similar way. Using the information they have on “similar people”, they dig through their large datasets and produce items which they think you might also enjoy. Even the ads and “people you might know” you see on Facebook are discovered in this way.
Feizi’s algorithm provides a simple mathematical way to fine tune these recommendations by filtering out the indirect, transient connections from the truer, more meaningful ones. It could turn out to be quite a powerful mechanism, not only when applied to biological processes (e.g. drug effects on gene expression) but also to social networks such as Twitter and LinkedIn, recommendation engines, translation tools, and trust metrics.
The Broad Institute have published an article which summarises the research nicely.
Feizi, S., Marbach, D., Médard, M., & Kellis, M., Nature Biotechnology, 31, 726 (link)