Toward Automatically Drawn Metabolic Pathway Atlas with Peripheral Node Abstraction Algorithm

Toward Automatically Drawn Metabolic Pathway Atlas with Peripheral Node Abstraction Algorithm

Myungha Jang, Arang Rhie, and Hyun-Seok Park*

Bioinformatics Laboratory, School of EngineeringEwha Womans University

Seoul, Korea

IEEE BIBM, 18-21 Dec 2010, Hong Kong Ewha Womans University

Table of Contents


I. Introduction

II. Topological Nature of Metabolic Networks at Peripheral Nodes

III. Node Abstraction Featured Scale-free Algorithm

IV. Experimental Results

V. Discussion and Future Work

• Abstract graph structure ⇒ visual representation

• Graphical diagrams are intuitively helpful to understand

biochemical reaction networks

- Node : compound, Edge : reactions

•Optimal solutions : NP-hard problems

Automatic graph layout algorithms in systems biology


I. INTRODUCTION

• A complete metabolic network indicates all the metabolic potential and capacity. • The shift of research focus: single pathways to multiple pathways. • Visualization serves an important role in understanding large scale metabolic

network.• KEGG Atlas(http://www.genome.ad.jp/kegg), 2008 • Terms : Global (metabolic) pathway, Multiple pathway, Atlas

Focusing on Global Metabolic Pathway


I. INTRODUCTION

http://www.genome.ad.jp/kegg

Our Efforts Toward Automatic Global Layout


I. INTRODUCTION

• Not enough to deal with the global pathway!• How can we obtain a complete view? • No attempts for automatic visualization for Atlas

Related work: KEGG Atlas • The map integration process is carried out manually by curators. • Based on curator’s experience • However, that metabolic networks are dynamic in nature should not be

disregarded Systematic approach is necessary

How To Deal With Large-scale Metabolic Pathway?I. INTRODUCTION


How To Deal With Large-scale Metabolic Pathway? (con’d)INTRODUCTION

Our Strategy

We provide a novel algorithmic approach in drawing multiple metabolic pathways by considering two properties:

1. Automatic abstraction criteria: by analyzing a topological nature of metabolic networks based on the graphical property of relation distance, linear reactions were abstracted as a unit reaction.

2. the consistency of highly connected nodes

• We obtained 255 map data by parsing KEGG XML (KGML) documents of version 0.6 using our KGML Parser.

Two terms were defined: 1. Relation degree the number of edges branching from a node 2. Relation distance a factor to measure the length between any two compounds encompassing nodes which all have relation degrees less than or equal to p (p = 2)

KGML

+

• A dedicated analysis on peripheral nodes with low connectivity was performed.


II. TOPOLOGICAL NATURE OF METABOLIC NETWORKS AT PERIPHERAL NODES

Relation Distance Term Clarification


• Definition: The length between any two compounds encompassing nodes which all have relation degrees equal to p • Here, p = 2


Relation Distance Example in Map


cpd:C01291

cpd:C01290

cpd:C16475

cpd:C16466

cpd:C16470

cpd:C16468

cpd:C16469

cpd:C16471

cpd:C00369

RD(C01290, C00369) = 7


Layout Components according to High Connectivity

Basic Motivation

• Observation: 66.83% of the total compounds within the complete metabolic pathways were of low connectivity, with less than relation degree of 3.• The number of compounds with higher relation degree, i.e. more than 6 edges, was much less.

Abstracting Compounds With Linear Interaction


III. NODE ABSTRACTION FEATURED SCALE-FREE ALGORITHM

A. Abstracting Compounds With Linear Interaction

• We abstracted and hid all those compounds that appear within these linear interactions. • This approach could be called “chain reduction”(M. Chimani et al) • All green compounds in the figure will be hidden in the graph layout according to this approach.



B. Layout Components according to High Connectivity

Input : Metabolic Pathway Graph

Output : coordinates of each node

void LayoutPathway (Pathway graph)

{

IF highly connected nodes (Nd) exist in graph

LayoutHighlyConnectedNode (graph, Nd);

ELSE IF any cycle(Nc) exists in graph

AND size of cycle ≥ 6

LayoutCircular (graph, Nc);

ELSE LayoutHierarchic (graph);

}

• Highly Connected Nodes: Nodes with relation degree bigger than 6

• LayoutHighConnectedNode() Algorithm Steps1. Find a highly Connected node Nd

2. Each component connected to Nd is decomposed into sub-graph

3. Each decomposed sub-graph is treated as a super node to apply the spring-embedding algorithm

63



IV. EXPERIMENTAL RESULTS

Experiments : To compare compression rate of compounds, we obtained the number of

abstracted compounds and edge crossings by applying two different layout algorithms:

Result 1

• Node compression rate performance

• Scope

1. 84 single metabolic pathways

2. 8 major categorized metabolic pathways

3. the global pathway

Result 2

• The number of edge crossing comparison between by

1. Conventional algorithm

2. Our Node abstraction featured scale-free layout algorithm


…

single pathways

… Categorized pathways

Globalpathway

III. EXPERIMENTAL RESULTS

Pathway Number of Nodes Before Abstraction

Number of Nodes After Abstraction Abstraction Rate

Carbohydrate Metabolism 1235 972 21.2%

Lipid Metabolism 1043 805 22.8%

Nucleotide Metabolism 424 351 17.21%

Amino Acid Metabolism 1327 980 26.14%

Metabolism of Other Amino Acid 332 262 21.08%

Metabolism of Cofactor and Vitamins

250 175 30%

Biosynthesis of Secondary Metabolism

800 536 33%

Xenobiotics Biodegradation 542 348 35.79%

Global Pathway (Atlas) 5675 4371 22.98%

Result 1B

The Number of Nodes Before and After Applying Node Abstraction


Peripheral path as supplementary nodes


Results drawn with Cytoscape, using conventional spring embeddingThe red-colored edges represent the abstracted edges. (abstraction rate : 70%)

Result 1A


Original Network Abstracted Network

Peripheral path as super edges


• In single metabolic pathways, the node abstraction featured algorithm

reduced edge crossings by 63.31%.

• In a global metabolic pathway, the number of edge crossings has reached a

reduction of 58.08% in total.

• Our proposed algorithm with node abstraction resulted in 86,067 edge

crossings, whereas the one without node abstraction resulted in 205,316 edge

crossings.

Result 2 : Edge Crossing Reduction


IV. DISCUSSION

• Two approaches were used: 1. Abstracting compound pairs according to a consistent criteria 2. Layout components according to high connectivity

• Our experimental results show that node abstraction feature reduced the number of compounds by approximately 23% in global pathway.

• Further discussion is necessary regarding enzyme reactions


IV. WHY IS OUR WORK IMPORTANT?

• The first systematic approach for Atlas visualization focusing on peripheral nodes

• Fundamental to building a hierarchical structure of Atlas

• Our approach is flexible upon pathway database change that frequently updates

• It is a crucial preliminary step toward automatically drawn metabolic pathway

• Future research on individual biological meaning of each peripheral nodes and abstracted path


Toward Automatically Drawn Metabolic Pathway Atlas with Peripheral Node Abstraction Algorithm

Documents