MORO: a Cytoscape App for Relationship Analysis between Modularity and Robustness in Large-Scale Biological Networks Authors: Cong-Doan Truong , Tien-Dzung Tran and Yung-Keun Kwon 1 October 8, 2016
Contents l Motivation l Modularity and robustness definition l Implementation & Results -Analysis of modularity and robustness -Module visualization -Module centrality & GO analysis -Parallel computation of robustness l Conclusions 2
Motivation High modularity (0.40518) Low modularity (0.08654) • Dynamical behaviors, particularly robustness, of biological networks can be highly affected by their modularity characteristics • MORO is Cytoscape app for analyzing relationship between modularity and robustness Complex Systems Computing Lab 3
Modularity and robustness definition Complex Systems Computing Lab 4
Modularity definition Ø Given a directed graph 𝐻(𝑊, 𝐹) 𝑄 = {𝑊 * ,𝑊 + } • Module V 1 : • G 𝑥 8 9 : {𝐵 → 𝐶, 𝐵 → 𝐷, 𝐶 → 𝐷} • A F >?@ : : {𝐷 → 𝐸} 𝑥 8 9 • C D B BC : : {𝐻 → 𝐷} 𝑥 8 9 • E Module V 2 : • 𝑥 8 E : {𝐸 → 𝐹, 𝐹 → 𝐺, 𝐺 → 𝐻, 𝐻 → • Module V 1 Module V 2 𝐸, 𝐸 → 𝐺, 𝐻 → 𝐹} 𝒋𝒐 𝒙 𝑾𝒋 𝒑𝒗𝒖 𝒙 𝑾𝒋 𝒙 𝑾𝒋 >?@ : : {𝐻 → 𝐷} 𝑵 𝑥 8 E • 𝑵 𝑸 = ∑ ( 𝒙 − ) ∈ [0, 1] 𝒋S𝟐 𝒙 𝟑 𝑵 𝑯 = 𝒏𝒃𝒚 𝑸 𝑵(𝑸) BC : : {𝐷 → 𝐸} 𝑥 8 E • • Leicht EA, Newman MEJ: Community Structure in Directed Networks . Physical Review Letters 2008, 100 (11):118703 ^ *∗* a *∗* 𝑁 𝑄 = ** − ** E + ** − ** E = • • Noack A: Modularity clustering is force-directed layout . Physical Review E 2009, 79 (2):026102 𝟏.𝟗𝟏𝟐𝟕𝟔 Complex Systems Computing Lab 5
Network robustness definition original attractor 𝒕 𝑤 t 𝑤 ^ Initial state OR OR Inhibit 01001111 01001100 11000010 𝑤 a 𝑤 + 𝑤 u 𝑤 u Initial state new attractor 𝒕 𝒘 𝒋 AND AND Activate OR 0100111 0 01001010 𝑤 v 𝑤 * v 0 𝑤 w AND AND OR C 1 𝑜|𝑇|k k 𝐽( 𝑡 = 𝑡 n o Robustness : 𝛿 𝐻 = p ) q∈r BS* Complex Systems Computing Lab 6
In-/Out- module robustness definition 𝑁𝑝𝑒𝑣𝑚𝑓 𝑊 • In-Module robustness + 𝑁𝑝𝑒𝑣𝑚𝑓 𝑊 * 1001100110 1001100110 0011100110 G G 1101100110 1010101110 0001101110 A D 100 001 F C Calculation of attractor 000 B B similarity E • Out-Module robustness 𝛿 BC (𝑊 * ) H 1001101110 0101001011 I 1100110 1100110 Calculation of J attractor similarity 1001011 • 𝛿 BC 𝐻 = * z z ∑ 𝛿 BC 𝑊 𝛿 >?@ (𝑊 * ) B BS* • 𝛿 >?@ 𝐻 = * z 𝑁𝑝𝑒𝑣𝑚𝑓 𝑊 z ∑ 𝛿 >?@ 𝑊 ^ B BS* Complex Systems Computing Lab 7
Implementation & Results 1. Case study 2. Module visualization 3. Module centrality & GO analysis 4. Parallel computation of robustness Complex Systems Computing Lab 8
Signaling networks • STKE network: – consists of 754 genes and 1,624 interactions – http://stke.sciencemag.org • HSN network: – The human signal transduction network (www.bri.nrc.ca/wang) – consists of 5,443 genes and 37,663 interactions The canonical cell signaling network (STKE network) Complex Systems Computing Lab 9
Analysis of modularity and robustness • STKE network • HSN network • Modules: 16 • Modules: 22 • Modularity: 0.72825 • Modularity: 0.54534 • Robustness: 0.67721 • Robustness: 0.75400 Complex Systems Computing Lab 10
Random Boolean network (RBN) model • Shuffling interaction model • Barabási-Albert (BA) • Erdős-Rényi (ER) • Erdős-Rényi variant model Scale free network (BA) Erdős-Rényi (ER) Complex Systems Computing Lab 11
Analysis of modularity and robustness • Generate 6400 RBNs (BA model) and then examine the correlation between modularity and robustness 1 0.8 network robustness 0.6 0.4 0.2 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 network modularity random networks HSN STKE Correlation coefficient = −0.80303 with p-value < 10 −4 Complex Systems Computing Lab 12
Relationship of the network modularity to the in-/out- module robustness - Relationship of the modularity to the in- module robustness (R= −0.30383, p- value <10 -4 ). - Modularity and out- module robustness (not significant). - Network robustness to the in-module robustness (R = 0.27801, p-value <10 - 4 ). - Network robustness and out-module robustness (not significant). Complex Systems Computing Lab 13
Module visualization (1) • Detailed visualization mode • A brief mode with absolute relations • A brief visualization with relative mode Complex Systems Computing Lab 14
Module visualization (2) Ø Results of the detailed visualization mode 22 modules of HSN network 16 modules of STKE network Complex Systems Computing Lab 15
Module visualization (3) HSN STKE Absolute mode STKE HSN The reduced visualization results after removing all links except about 30% of links with the highest weight values Complex Systems Computing Lab 16
Module visualization (4) STKE HSN Relative mode STKE HSN The reduced visualization results after removing all links except about 30% of links with the highest weight values Complex Systems Computing Lab 17
Module centrality analysis How each module is positioned in terms of relations among the modules? • Five well-known centrality methods • Degree (DEG) • Closeness (CLO) • Betweeness (BEW) • Stress (STR) • Eigenvector (EIG) Complex Systems Computing Lab 18
Module centrality result Ø The correlation between five centrality values and module sizes of STKE network The module size which is defined as the number of nodes belonging to the module showed positive relationships with all module centrality measures Complex Systems Computing Lab 19
GO analysis Choose largest module Select the rest of module • http://www.uniprot.or g/ • http://www.ebi.ac.uk/ QuickGO/ Ø The interface of GO analysis function in MORO app Complex Systems Computing Lab 20
Parallel computation of robustness Ø Running time of MORO based on three modes such as single CPU, Multi-core CPU and GPU) with number of considered initial-states (1000) 10000 Running time (logarithimic 100 Running time (logarithimic 1000 scale based 10) 10 scale based 10) 100 1 10 0.1 1 0.01 Single CPU Multi-core CPU GPU Single CPU Multi-core CPU GPU Running mode Running mode STKE network HSN network Complex Systems Computing Lab 21
Some interfaces of MORO Cytoscape app Complex Systems Computing Lab 22
Conclusions • Summary : – Analyze the relationship between network robustness and modularity – Provide various module visualization modes – Analyze module centrality by employing five well-known methods – Analyze gene ontology of two groups of modules – Implement robustness algorithms in parallel – Provide a batch-mode simulation • Future works : – Consider various types of mutations such as a knockout and edge mutation – Extend Boolean network model by using ordinary differential equations (ODEs) Complex Systems Computing Lab 23
Thank you for your attention! Any question?
Recommend
More recommend
