CMStatistics 2016: Start Registration
View Submission - CMStatistics
B1325
Title: Statistical characterisation of diffusion-based approaches in biological networks Authors:  Sergio Picart-Armada - Universitat Politecnica de Catalunya (Spain) [presenting]
Francesc Fernandez-Albert - Takeda Cambridge (United Kingdom)
Wesley Thompson - Institute of Biological Psychiatry (Denmark)
Alfonso Buil - Mental Health Center Sct. Hans (Denmark)
Alexandre Perera-Lluna - B2SLab at UPC (Spain)
Abstract: Network analysis in computational biology pursues understanding experimental data in the context of known interactions between genes, proteins or metabolites, aiming to unravel new interactions, find molecular signatures or characterise new mechanisms. The analysis of diffusion processes in these networks quantifies how a perturbation starting at some seed nodes (e.g. downregulated genes) propagates to the rest of biological entities - ultimately identifying affected subnetworks. Diffusion-based approaches are robust to the noisy nature of experimental data and the presence of spurious associations in the knowledge model, but they are inherently related to the topology of the network. We analyse the statistics of diffusion scores through the introduction and characterisation of null models. We explore (1) the influence of topology, (2) the impact that the distribution of seed nodes has on the diffusion scores, (3) possible hypotheses testing on single nodes or whole subnetworks and (4) the definition of biologically sound null models. We find that the network architecture leverages the diffusion states and that the topological features of the nodes are reflected in their null distribution. Furthermore, nodes with correlated null distributions are prone to share biological functions. On the other hand, the success of hypotheses testing or subnetwork selection also depends on the seed nodes distribution and the formulation of the null model.