Complex Networks ________________________________________________

Soumen Roy

Research in Complex networks at Bose Institute is focussed in two directions: (i) the study of networks in different biological systems and contexts, and, (ii) various fundamental problems in network  theory (network growth models,  network metrics and exploiting them meaningfully in a given system, ...), which have important consequences not just for biological networks but for all networked systems in general.

Drawing on inputs from  collaborators in various experimental labs, we construct (using experimental data, results from text-mining, ...) and analyse networks in different biological systems (microbes, plants, ...), with a keen eye towards identifying  biologically important entities. One of the questions that we specifically seek to answer is: (how) does network topology encode biological phenotype?

Microbe phenotypes and their network architecture

We use an automated approach to characterize metabolic networks of 32 microbial species using 11 topological metrics from complex networks. Clustering allows us to extract the indispensable, independent, and informative metrics. Using hierarchical linear modeling, we identify relevant subgroups of these metrics and establish that they associate with microbial phenotypes surprisingly well. This work can serve as a stepping stone to cataloging biologically relevant topological properties of networks and toward better modeling of phenotypes. The methods we use can also be applied to networks from other disciplines.

Further details: Strong associations between microbe phenotypes and their network architecture, Physical Review E (Rapid Communication), 80, 040902 (2009)

Modeling and verifying a broad array of network properties: graphlet growth

Almost all papers in network theory focus on only one or two  network metrics at a time. Very natural questions arise: (a) why shouldn't all  network metrics be studied together instead of studying just one or two, (b) why should'nt higher moments of metrics be studied, (c) how to identify, which of these metrics are most informative (or redundant), in any given situation? And finally, given all this information, what systemic or emergent properties of the system(s) under study can we extract.

Motivated by the observation that network growth often takes place via addition of graphlets rather than single nodes (modules in  biology and software, families in social networks etc) , we attempt to answer the above questions.

Linear independence of network metrics
Network PCA analysis

(Top) Heatmap showing linear independence of network metrics (Bottom) coverage of PCA space by the graphlet growth model (orange), Barabasi-Albert model (blue) and 113 real-world networks (black dots).

Further details: Modeling and verifying a broad array of network propertiesEurophysics Letters 86 (2009) 28003