• People
  • Web mail
  • Internal and Reserved Area
  • Print

Prof. Stefano Boccaletti


Tuesday, March 19th, 2013 at 02:30:00 PM  

CNR (building F), room 2, via Madonna del Piano 10, Sesto Fiorentino

Published on-line at 06:22:45 PM on Thursday, February 28th, 2013

Parencletic networks representation of datasets

Third seminar in the series "Sguardi sulla complessità" on complexity by CSDC, ISC and Physics Dept. of CNR and Università di Firenze.

Of the different ways of representing a complex system, the one afforded by networks is certainly among the richest and most general. Endowing a system with a network representation requires defining nodes and links connecting them. Often physical or virtual relationships between the elements of the system, e.g. anatomic brain fibres or hyperlinks between the pages of a website, constrain the way a link is defined. In the absence of such relationships, functional links can still be built, but this is only possible when a vector of observables can be associated to each node, e.g. the time evolution of a stock price, or of brain activity in a given region. To overcome this limitation, we propose a novel method which allows treating collections of isolated, possibly heterogeneous, scalars, e.g. sets of biomedical tests, as networked systems. The method builds a network where each node represents an observable, and links codify the distance between a pair of observables and a model of their typical relationship within the studied population.

Topological characteristics can then be used to extract important information from the system. In particular, atypical or pathological conditions correspond to scale-free networks, whose hubs are the elements that best explain them, whereas typical or normative conditions are characterized by sparsely connected networks with homogeneous nodes. Insofar as a network representation of each instance or subject is constructed with reference to the population to which he is compared, this method is by its very nature a difference seeker.

We use our method to analyze the response of the plant Arabidopsis Thaliana to hosmotic stress, using genetic expression levels as observables. The most important genes turned out to be the nodes with highest centrality in the reconstructed networks. Knocking out these genes resulted in phenotype expression rates one order of magnitude higher than equivalent screenings. Finally, our method not only confirmed known results, but also highlighted important genes hitherto unrelated to the hosmotic stress response.