Final Report Summary - GRODYNET (Group-based Dynamics on Complex Networks)
Thanks to this project we have advanced considerably in our understanding of the group-structure of complex networks, and in the development of inference methods that use group-based models to predict the evolution and the dynamics of networks.
We have applied these approaches to social and biological systems. With regard to social systems, we have demonstrated that it is possible to anticipate human behaviors. For example, we can use network representations to predict the decisions of justices in a Court, or whether someone will like a certain movie or not. We have also conducted an empirical study and shown that we can predict the appearance and resolution of personal conflict in small teams of people. In all cases, node groups (groups of justices, of movies, or of team members) hold the key to accurate predictions. In fact, despite not taking into account domain knowledge (for example, about the judicial system in the case of justices), our inference methods are more accurate than domain-specific methods in the literature.
We have also made progress in the application of these tools to biological problems. First, we have found that the relevant dynamics in cellular systems occurs at the level of groups of nodes with similar connection patterns rather than individual nodes. As we anticipated, this is true and relevant for the understanding of cell polarity in yeast. Indeed, we have been able to uncover the group structure of the network of proteins involved in the process, and demonstrated that a model with a single parameter for each group of proteins predicts the dynamics of individual proteins.
We have also studied the dynamics of cell migration and demonstrated that, despite the complex network of protein interactions in the cellular adhesome (the system in charge of transmitting forces between cells), two specific proteins are highly predictive of cellular phenotypes. The predictive models we have developed are important, among others, for understanding cell migration of cancer cells and metastasis.
Finally, we have shown that it is possible to predict interactions between drugs using network approaches; again, the key is that there are groups of drugs that have similar behaviors and exploiting this information at the large scale. Predicting interactions between drugs is important to avoid potentially harmful combinations, to reduce off-target effects of treatments and to fight antibiotic resistant pathogens, among others.
Roger Guimerà holds a permanent ICREA Research Professor position at Universitat Rovira i Virgili. More information about him and the project can be found at http://seeslab.net/grants/group-based-dynamics-complex-networks/ or cat roger.guimera@urv.cat.