10/4/2012. Analysis and Visualization of Biological Networks with Cytoscape. Outline of the Day. Introductions

Transcription

1 Analysis and Visualization of Biological Networks with Cytoscape John Scooter Morris, Ph.D., UCSF October 4, Outline of the Day Introductions and setup (15 minutes) Biological Networks (60 minutes) Analytical Approaches Visualization Break Introduction to Cytoscape (45 minutes) Guided tutorial Introduction to Cytoscape II (45 minutes) Lunch Plugin Demos and Use Cases (25 minutes) Tips & Tricks (15 minutes) Working with Data (25 minutes) Analysis with Microarray Data (25 minutes) Break In-depth Examples with Your Data Future directions Cytoscape 3.0 Questions and (hopefully) Answers 2 Introductions John Scooter Morris Currently Adjunct Assistant Professor, Pharmaceutical Chemistry Director, NIH Resource for Biocomputing, Visualization, and Informatics UCSF Roving Engineer, NIH National Resource for Network Biology (NRNB) Distinguished Systems Architect: Genentech, Inc. Cytoscape core team since 2006 Author of several Cytoscape plugins SFLDLoader, structureviz, clustermaker, chemviz, metanodeplugin, grouptool, commandtool, biocycplugin 3 1

2 Introductions Alex Pico, Ph.D Current Executive Director, National Resource for Network Biology 2007-Current Bioinformatics Group Leader, Gladstone Institutes, UCSF Cytoscape core team since 2007 Developer of several Cytoscape plugins: CriteriaMapper, GenMAPP Workspaces, GO-Elite, GOLayout, BubbleRouter, Mosaic. 4 Introductions You? Biologists Bioinformaticists Mathematicians Computer Scientists Molecular Biogists Chemists Other? 5 Introductions What do you hope to get from today s workshop? 6 2

3 Installation CD Contents Cytoscape installers for Mac, Linux, Windows Several additional plugins Sample data files PDFs for hands-on portions 7 Installation If you have not yet installed Cytoscape 2.8: Install Cytoscape by executing the appropriate installer If desired, copy plugins into your Cytoscape plugins folder Linux: {Cytoscape install directory}/plugins Included in installers! Windows: \Program Files\Cytoscape_v2.8.0\plugins Mac OS X: /Applications/Cytoscape_v2.8.0/plugins Alternatively, can use Install Plugin From File to install the desired plugins 8 What is a network tool? Networks provide an integrating context for data Commonly understood diagrammatic representation for concepts and relationships Provides structure that helps reduce underlying complexity of the data More efficient than searching databases element-by-element Network tools give us functionality for studying complex processes Analysis of global characteristics of the data, e.g. degree, clustering coefficient, shortest paths, centrality, density Identify key elements (hubs) and interesting subnets Help elucidate mechanisms of interaction Visualization of data superimposed upon the network Help understand how a process is modulated or attenuated by a stimulus 9 3

4 10/4/2012 Applications of Network Biology jactivemodules, UCSD Gene Function Prediction shows connections to sets of genes/ proteins involved in same biological process Detection of protein complexes/ other modular structures discover modularity & higher order organization (motifs, feedback loops) Network evolution biological process(s) conservation across species Prediction of new interactions and functional associations Statistically significant domaindomain correlations in protein interaction network to predict protein-protein or genetic interaction mcode, University of Toronto DomainGraph, Max Planck Institute PathBlast, UCSD 10 Applications of Network Informatics in Disease Agilent Literature Search Identification of disease subnetworks identification of disease network subnetworks that are transcriptionally active in disease. Subnetwork-based diagnosis source of biomarkers for disease classification, identify interconnected genes whose aggregate expression levels are predictive of disease state Subnetwork-based gene association map common pathway mechanisms affected by collection of genotypes (SNP, CNV) PinnacleZ, UCSD Mondrian, MSKCC 11 Biological Networks Motivation/Challenge Biological Network Taxonomy Network metrics and analysis Concepts Scale-free networks Random networks Network measures Clustering Network motifs Overrepresentation analysis 12 4

5 10/4/2012 The Challenge Making sense out of biological networks. 13 The Challenge Biological networks (nodes and edges) Seldom tell us anything by themselves Analysis involves: Understanding the characteristics of the network Modularity Comparison with other networks (specifically random networks) Visualization involves: Placing nodes in a meaningful way (layouts) Mapping biologically relevant data to the network Node size Node color Edge weights 14 The Challenge 15 5

6 10/4/2012 Biological Network Taxonomy Pathways Signaling Metabolic Regulatory Phylogeny (could also be thought of as similarity) 16 Biological Network Taxonomy Interactions Protein-Protein interactions Protein-Ligand interactions Genetic interactions Domain-Domain interactions Others Residue or atomic interactions Cell/cell interactions Population biology Epidemiology Social networks 17 Biological Network Taxonomy Similarity Protein-Protein similarity Chemical similarity Ligand similarity (SEA) Others Tag clouds Topic maps 18 6

7 Analytical approaches Concepts Graph/Network correspondence Node/Vertex correspondence Edge directedness Usually a network property Node degree Multigraph Allow multiple edges between nodes Hypergraph Allow edges to connect more than 2 nodes 19 Analytical approaches Scale-free networks Degree distribution follows power law: P(k) ~ k -γ, where γ is a constant. Result is that there are distinctive hubs (essential proteins?) Overall, though, network is resilient to perturbation Biological (and social) networks tend to be scalefree 20 Analytical approaches Random networks Algorithms exist to create random networks Flat random network: Erdos-Renyi High clustering coefficient: Watts-Strogatz Scale-free: Barabasi-Albert Nodes have similar degrees Useful to compare your network vs. a random network 21 7

8 Analytical approaches Network measures Node degree Node indegree: # of edges for which this node is a target Node outdegree: # of edges for which this node is a source Shortest path length Shortest traversal distance between two nodes Can be weighted if edges have weights or hops if not 22 Network measures Node degree Analytical approaches Node indegree: # of edges for which this node is a target Node outdegree: # of edges for which this node is a source Shortest path length.5 Shortest traversal distance between two nodes.3 Can be weighted if edges have weights or hops if not Clustering coefficient Measures how close the neighbors of a node are to being a clique (fully connected group) 0 # of edges connecting a node s neighbors/the node s degree 1 23 Analytical approaches Centrality - measures of node importance Degree centrality (find hubs) Degree of this node / (# of nodes 1) Betweenness centrality (essentiality) The average number of shortest paths that go through this node Closeness centrality The sum of all shortest paths between this node and all other nodes / (# of nodes 1) 24 8

9 10/4/2012 Analytical approaches Clustering (find hubs, complexes) Goal: group related items together Clustering types: Hierarchical clustering Divide network into pair-wise hierarchy K-Means clustering Divide network into k groups MCL Uses a flow simulation to find groups Community Clustering Maximize intra-cluster edges vs. inter-cluster edges 25 Analytical approaches Motif finding Search directed networks for network motifs (feedforward loops, feedback loops, etc.) Overrepresentation analysis Find terms (GO) that are statistically overrepresented in a network Not really a network analysis technique Very useful for visualization 26 Analytical Approaches Network Analyzer Demo 27 9

10 10/4/2012 Visualization of Biological Networks Depiction Data Mapping Layouts Animation 28 Depiction There are various ways to depict biological networks: Node-Link (graph) representation This is what we most often think of Partitioned Node-Link representation Split graph into discrete partitions based on some feature Matrix representation Can be useful for very dense networks Can also map information into cells of matrix e.g. degree, color scale (heat map) 29 Data mapping Mapping of data values associated with graph elements onto graph visuals Visual attributes Node fill color, border color, border width, size, shape, opacity, label Edge type, color, width, ending type, ending size, ending color Mapping types Passthrough (labels) Continuous (numeric values) Discrete (categories) 30 10

11 10/4/2012 Data mapping Avoid cluttering your visualization with too much data Map the data you are specifically interested in to call out meaningful differences Mapping too much data to visual attributes may just confuse the viewer Can always create multiple networks and map different values 31 Layouts Layouts determine the location of nodes and (sometimes) the paths of edges Types: Simple Grid Partitions Hierarchical layout data as a tree or hierarchy Works best when there are no loops Circular (Radial) arrange nodes around a circle could use node attributes to govern position e.g. degree sorted 32 Layouts Types: Force-Directed simulate edges as springs may be weighted or unweighted Combining layouts Use a general layout (force directed) for the entire graph, but use hierarchical or radial to focus on a particular portion Multi-layer layouts Partition graph, layout each partition then layout partitions Many, many others 33 11

12 Layouts Use layouts to convey the relationships between the nodes Layout algorithms may need to be tuned to fit your network Layoutsà Settings menu Lots of parameters to change layout algorithm behavior Can also consider laying out portions of your network 34 Animation Animation is useful to show changes in a network: Over a time series Over different conditions Between species 35 Introduction to Cytoscape Overview Core Concepts Networks vs. Attributes VizMapper Plugins Apps Working with Data Loading network from the Web Importing networks from csv files or Excel Importing attributes from csv files or Excel The attribute browser Cytoscape tips & tricks 36 12

13 Cytoscape About Cytoscape: Cytoscape is an open source bioinformatics software platform for visualizing molecular interaction networks and integrating these interactions with gene expression profiles and other state data. Cytoscape Web Site 37 Visualization Integration Analysis WHAT IS CYTOSCAPE? 38 What is Cytoscape? Open source Cross platform A consortium University of Toronto University of Michigan Unilever 39 13

14 Core Concepts Networks and Annotations Networks e.g, biological pathways Annotations e.g., attributes or data 40 Core Concepts Visual Mapping with VizMapper Networks VizMapper Annotations 41 Visual Styles Use specific line types to indicate different types of interactions Encode specific physical entities as different node shapes. Set node sizes based on the degree of connectivity of the nodes. Browse extremely dense networks by controlling for the opacity of nodes Set node font sizes based on the degree of connectivity of the nodes. 42 Expression data 14

15 Core Concepts Apps! 43 Cytoscape Traditional uses Visualizing: PPI Pathways Mapping: Expression profiles Other state data Analysis: Network properties Data mapped onto network 44 Loading Networks Use import network from table: Excel file Comma or tab delimited text 45 15

16 Loading Attributes Loading attributes Use import attribute from table 46 Loading Attributes Loading attributes Use import attribute from table The Data Panel 47 Cytoscape Sessions In Cytoscape Open and Save mean Open Session and Save Session Sessions save pretty much everything: Networks Properties Visual styles Screen sizes 48 16

17 10/4/2012 Tutorial #1 Introduction to Cytoscape Part 1 Tutorial:Introduction_to_Cytoscape 49 Tutorial #2 Introduction to Cytoscape Part 2 Tutorial:Introduction_to_Cytoscape-part2 50 Cytoscape: Platform Cytoscape as a platform Plugin architecture BINGO GO Over expression analysis MCODE Finding protein complexes Agilent literature search plugin Creating networks from the literature 51 17

18 Examples/Demos clustermaker Clustering and cluster visualizations BiNGO Gene Ontology overrepresentation analysis structureviz Bringing protein structure to networks Mosaic Detangling network hairballs into pathways 52 Tips & Tricks Root graph There is one graph to rule them all. The networks in Cytoscape are all views on a single graph. Changing the attribute for a node in one network will also change that attribute for a node with the same ID in all other loaded networks There is no way to copy a node and keep the same ID Make a copy of the session 53 Tips & Tricks Network views When you open a large network, you will not get a view by default To improve interactive performance, Cytoscape has the concept of Levels of Detail Some visual attributes will only be apparent when you zoom in The level of detail for various attributes can be changed in the preferences To see what things will look like at full detail: Viewà Show Graphics Details 54 18

19 Tips & Tricks Network views When you open a large network, you will not get a view by default To improve interactive performance, Cytoscape has the concept of Levels of Detail Some visual attributes will only be apparent when you zoom in The level of detail for various attributes can be changed in the preferences To see what things will look like at full detail: Viewà Show Graphics Details 55 Tips & Tricks Some really useful commands Select the neighbors of a node Selectà Nodesà First Neighbors of Selected Nodes Finding the nodes on either side of an edge Selectà Nodesà Nodes Connected by Selected Edges Search for a node based on an attribute value: Click on the tools icon ( ) next to the search box Select the attribute you want to search under Select Attributeand click Apply The Search box will now search within the attribute you specified 56 Tips & Tricks Some more really useful commands The icon is a short-cut to doing a Force-directed layout To do a layout of only some nodes, select the nodes and then do Layoutà Cytoscape Layoutsà Force-Directed Layoutà Selected Nodes Only To change the background color (or any default visual property) Open the VizMapper tab and click on the two-node network labeled Defaults The background color is in the Global tab To force node height and width to be equal: Go to the default appearance dialog Click on dependencies 57 19

20 Tips & Tricks Logging By default, Cytoscape writes it s logs to the Error Dialog: Helpà Error Dialog Can change a preference to write it to the console Edità Preferencesà Properties Set logger.console to true Don t forget to save your preferences Restart Cytoscape (can also turn on debugging: cytoscape.debug, but I don t recommend it) 58 Tips & Tricks Memory Cytoscape uses lots of it Doesn t like to let go of it An occasional restart when working with large networks is a good thing Destroy views when you don t need them Java doesn t give us a good way to get the memory right at start time Cytoscape 2.7 does a much better job at guessing good default memory sizes than previous versions 59 Tips & Tricks.cytoscape directory Your defaults and any plugins downloaded from the plugin manager will go here Sometimes, if things get really messed up, deleting (or renaming) this directory can give you a clean slate Plugin manager Outdated doesn t necessarily mean won t work Plugin authors don t always update their plugins immediately after new releases Click on Show outdated plugins to see the entire list of plugins

21 Getting Help 61 Tutoral Working with Data 62 Tutoral Analysis of Microarray Data 63 21

22 In-depth Examples Work through an example use case provided by the audience 64 Future Directions Cytoscape 3.0 Re-architecture of Cytoscape Significantly more modular (uses OSGi) Separation of UI from core algorithms Multiple roots Provides for better isolation between apps Provides ability for apps to export an API Beta release October 1, 2012 Final release December, Future Directions Cytoscape 3.0 beta demo 66 22

23 Future Directions RBVI Plugins chemviz: Improved renderer Support CyCommands (2.8)/API (3.0) Support for multiple fingerprints Support for determining maximum common substructure Support for substructure searching Expose additional chemical descriptors Cytoscape 3 version structureviz: Ability to create networks from structures residue interactions domain interactions Ability to interact with Cytoscape from within Chimera Cytoscape 3 version clustermaker: Fuzzy clustering Additional algorithms Cytoscape 3 version 67 Demo/Sample use cases Use case 1: Expression data anlaysis Plugins: BiNGO, clustermaker Use case 2: Protein complexes in protein-protein interaction networks Plugins: clustermaker 68 Expression data analysis Load galfiltered.cys Explore the expression fold changes: gal1rgexp, gal4rgexp, and gal80rexp. The network is colored by gal4rgexp values. To explore the expression profile for these three deletions, we can use clustermaker to do a hierarchical cluster Pluginsà Clusterà Hierarchical Choose the attributes we re interested in (node.gal1rgexp, node.gal4rgexp, node.gal80rexp) Choose the type of clustering (pairwise average-linkage, Euclidean distance) Click Create Clusters 69 23

24 Expression data analysis Now, visualize the clusters by clicking on Visualize Clusters This will bring up the TreeView If the colors are too dark, they can be adjusted in the Settings dialog. Select the dendrogram branch at the top, as shown Note that this will also select those proteins in the network view 70 Expression data analysis Now we can see if these genes show any GO overrepresentation Start BiNGO Pluginsà Start BiNGO Give the clusters a name (e.g. cluster1) Click on Start BiNGO to process the data The results include a table and a network of GO associations In this case the top term is galactose catabolic process via UDP-galactose, which makes good sense 71 Protein Complexes Load collinsplus.cys Three networks, but only one view combined_scores_good.txt: Combined MS/TAP Yeast PPI network from Collins, et. al. DNA and Tran b.csv: Yeast EMAP RNAPuberNov2+Meg6c.csv: Yeast EMAP 72 24

25 Protein Complexes Load collinsplus.cys Three networks, but only one view combined_scores_good.txt: Combined MS/TAP Yeast PPI network from Collins, et. al. DNA and Tran b.csv: Yeast EMAP RNAPuberNov2+Meg6c.csv: Yeast EMAP Find MCL clusters in the PPI network Pluginsà Clusterà MCL cluster Density Parameter: 1.8 Weak EdgeWeight Pruning Threshold: 1.0E-10 Maximum residual value: Array source: PE Score Click on Visualize Clusters 73 Protein Complexes Now that we have PPI clusters, we can explore some EMAP data Select the DNA and Tran network Now perform a hierarchical cluster Pluginsà Clusterà Hierarchical Cluster Click on Visualize Clusters to bring up the TreeView 74 Protein Complexes Now that we have PPI clusters, we can explore some EMAP data Select the DNA and Tran network Now perform a hierarchical cluster Pluginsà Clusterà Hierarchical Cluster Click on Visualize Clusters to bring up the TreeView Select your PPI clustered network in Cytoscape Search for GIM5 and select the entire cluster Note how the proteins in TreeView highlight Note that BUD27 doesn t cluster very well in EMAP Deselect BUD27 edge 75 25