The Role of Network Science in Biology and Medicine Tiffany J. Callahan Computational Bioscience Program Hunter/Kahn Labs Network Analysis Working Group 09.28.2017
Network-Enabled Wisdom (NEW) empirically derived networks are necessary for describing the molecular mechanisms and biological processes that drive disease under the influences of inherited risk factors (genetic markers) and environmental risk factors. Schadt EE, Björkegren JL. Science translational medicine. 2012
Outline Network Science Refresher Network Representation Data Integration Network Analysis
Expectations What this talk is not: A review of every aspect and domain of network science A comprehensive overview of network biology or network medicine Designed to make recommendations for network inference methods What this talk is: A general discussion of basic network science as it applies to specific problems in biomedical informatics A good starting place for those who want to implement network methods in biology or biomedical informatics
Network Science Refresher
What s a Network? G = (V, E) Sets of points connected by lines Points: nodes, vertices, people, biological entities Lines: edges, links, relationships, interactions Points (V) Lines (E) Why are networks useful? Mathematical representation Provide structure to complex data https://www.cc.gatech.edu/~dovrolis/courses/netsci/network-image.jpg Clauset CSCI 5352, Fall 2017
Network Types Single edge between pair of nodes No self-loops Undirected edges No node or edge annotations Simple Network Clauset CSCI 5352, Fall 2017
Network Types Multigraph Network Bipartite Network Directed Network Clauset CSCI 5352, Fall 2017
Network Types Temporal Network Planar Network Multiplex Network Hypergraph Clauset CSCI 5352, Fall 2017; Karimi et al., Physica A: Statistical Mechanics and its Applications; 2013
Network Characteristics Zhang P et al., Journal of molecular biology. 2017
Biological Networks https://www.slideshare.net/raunakms1/systems-biology-approaches-to-cancer-32303042
Network Representations
Network Representations Three main ways: Adjacency matrix Simple, symmetric Adjacency list Similar to a hash table Edge list Annotations and metadata Data integration Clauset CSCI 5352, Fall 2017
Biological Networks Nodes: genes, proteins, drugs, chemicals, organisms Edges: regulation, targets, interactions, reactions Marimuthu et al. J Proteomics Bioinform; 2011
https://www.slideshare.net/raunakms1/systems-biology-approaches-to-cancer-32303042
Disease Networks https://exploringdata.github.io/vis/human-disease-network/
Disease Networks Barabási. N Engl J Med; 2007
Data Integration
Why Integrate? https://www.slideshare.net/raunakms1/systems-biology-approaches-to-cancer-32303042
Integrating Heterogeneous Data Goal extract and combine knowledge from multiple datasets Challenges: Differing sizes and formats of data sources Dimensionality Complexity, noisiness Scalability Current Approaches: Projection methods Network propagation methods Kernel-based and Probabilistic methods REVIEW: Gligorijević V, Pržulj N. Journal of the Royal Society Interface. 2015
Network Analysis
Network Analysis Unsupervised or Exploratory Identify shape and pattern of the underlying network Post-hoc inductive hypothesis evaluation Benchmark à random graphs Examples: Missing edge prediction Supervised or Hypothesis-Driven Explain structure to get at mechanisms Domain specific and mathematically formalized Benchmark à gold standard Examples: Modularity or community detection Clauset CSCI 5352, Fall 2017
Human Diseasome Goal exploring whether human genetic disorders and the corresponding disease genes might be related to each other at a higher level of cellular and organismal organization. Approach Bipartite graph of diseases and genes Disease and gene connected by mutations OMIM; 1284 disorders and 1777 disease genes Results Essential human genes are hubs Most non-essential disease-genes are not hubs, instead localized to network periphery Diseases caused by somatic mutations are not peripheral confirmed with cancer genes Goh et al. Proceedings of the National Academy of Sciences. 2007
Goal Analyze network properties of disease networks in human interactome Community detection problem Approach Developed a new method that identifies disease modules given a set of proteins Used OMIM and GWAS studies Results Disease-related proteins do not tend to reside in dense local communities Method does as well if not better than Random Walk
Inferring new indications for approved drugs via random walk on drug-disease heterogeneous networks Goal Predicted new indications for approved drugs Approach Two-pass random walks with restarts on heterogeneous networks Drug-disease associations Drug-drug networks Disease-disease networks Compare their performance to 6 existing methods using 2 datasets Alzheimer s case study Results Correctly predicts drug-disease associations Performs as good or better than existing methods Predicted 9 out of 10 known Alzeimer s disease Liu et al. BMC bioinformatics. 2016
Conclusion & Acknowledgements Networks are awesome! Big Role in Biology and Medicine Relatively easy to generate Represent complex structures and mechanisms Facilitate heterogeneous data integration Exploratory and hypothesis-driven inference Supported by an active, diverse community Acknowledgements Dr. Aaron Clauset Jenny & Kim
Resources http://barabasi.com/networksciencebook/ http://www.tedmed.com/talks/show?id=7282 Clauset CU Boulder Class materials: http://tuvalu.santafe.edu/~aaronc/courses/5352/