Network diffusion-based analysis of genomic data.

The information on direct and indirect molecular interactions can be modelleddefining genome-scale networks. The integrated analysis of "-omics" measurementsand molecular interactions, referred to as network-based analysis, provides severalopportunities for a better interpretation of omics data, which is often hindered bybiological complexity and experimental biases. In comparison to theindependentanalysis of every statistical unit, network-based quantities interpret omics data tak-ing into account the modular and functional architecture of cells. Network-basedapproaches have been proposed in relation to several problems, including gene mod-ule identification, pathway analysis and patient stratification, justto mention a few.Among network-based approaches, the principle of spreading information through-out a network - namely network diffusion - has been recently used in several applica-tions, mainly related to the "smoothing" of sparse input quantities and the prioritiza-tion of molecular entities in network proximity. Here, we describe a network-diffusionbased framework for -omics data analysis aimed at identifying gene modules [1]. Thisframework is based on indices that jointly quantify molecular measurements and net-work location. The resulting ranked gene list is then analysed to assess the presenceof significant subnetworks. After having introduced the method -implemented as anR package named dmfind - and its performance in a controlled scenario, we presentthe results obtained on prostate cancer molecular profiles (somatic mutations andgene expression) and multiple gene lists associated with autism spectrum disorders.