Approximately Factorizable Networks
In factorizable networks, the adjacency (connection strength) between two nodes can be factored into node-specific contributions, named node “conformity”. Often the ith node conformity, CF i is approximately equal to the scaled connectivity k i ∕ sum(k) of the ith node. We describe (a) an algorithm for computing the conformity CF and for measuring the “factorizability” of a general network, and (b) a module- and CF-based decomposition of a general adjacency matrix, which can be used to arrive at a parsimonious description of a network. Approximately factorizable networks have important practical and theoretical applications, e.g., we use them to derive relationships between network concepts. Collaborative work with Jun Dong has shown that network modules (i.e., subnetworks comprised of module nodes) tend to be approximately factorizable (Dong and Horvath BMC Syst Biol 1(1):24, 2007).
KeywordsAdjacency Matrix Factorizability Function Module Assignment Conformity Vector Factorizable Network
- Gifi A (1990) Nonlinear multivariate analysis. Wiley, Chichester, EnglandGoogle Scholar
- Lange K (2004) Optimization. Springer, New YorkGoogle Scholar
- de Leeuw J, Michailidis G (2000) Block relaxation algorithms in statistics. J Comput Graph Stat 9:26–31Google Scholar
- Ranola JMO, Langfelder P, Song L, Horvath S, Lange K (2011) An MM algorithm for module and propensity based decomposition of a network. UCLA Technical ReportGoogle Scholar
- Servedio VDP, Caldarelli G, Butta P (2004) Vertex intrinsic fitness: How to produce arbitrary scale-free networks. Phys Rev E – Stat Nonlin Soft Matter Phys 70(5):056126Google Scholar