The document discusses using ontologies to integrate systems biology data. It describes typical steps in systems biology studies such as finding studies, processing data, integrating data, and combining data from multiple sources. Ontologies can help link information from different analysis techniques and combine data from many studies by capturing study metadata. The document advocates using standards like ISA-TAB and MAGE-TAB to capture study data and proposes using a generic study capture framework with modular components to integrate different types of 'omics data. Ontologies are needed for collaboration and to provide controlled vocabularies for annotation.

2. Using ontologies to do integrative
systems biology
Chris Evelo
Department of Bioinformatics - BiGCaT
Maastricht University
@Chris_Evelo
chris.evelo@maastrichtuniversity.nl

3. Typically we want to:
• Find studies.
• Process data.
• Integrate.
• Evaluate.
• Combine with yet
other data.
Faculty of Health, Medicine and Life Sciences

4. Systems Biology Issues:
• Environment
• Multi-compartment
• Different levels of gene expression cascade
(multi-omics)
Needs:
• Link information from different analysis
techniques
• Combine many studies (store study design)
Faculty of Health, Medicine and Life Sciences

5. Using ISA to
be able to
find studies
http://dx.doi.org/10.1038/ng.1054
Faculty of Health, Medicine and Life Sciences

6. Why a study capturing application?
New studies can be performed based on old data
Translational comparisons (mouse, human, rat etc)
Structured storage
Facilitate collaborations between groups
- Data sharing on joined project
- Start a collaboration

7. What do we need to accomplish this
Acceptance
- Using standards (e.g. ISA-TAB & MAGE-TAB)
- User friendly (interface via web browser)
- Open source
- Examples
Collaboration
- Ontologies
- Security of data (log-in and store data locally)
- Open source (make own module)

8. dbXP: a total study capturing solution
Simple assay module Metabolomics module
Web input Study capturing module Web output
Feature layer
Transcriptomics module Any new module

9. dbNP Architecture
GSCF Simple Assay module Query module
Body weight, BMI, etc.
Pathways, GO, metabolite profiles
Templates
Templates
Templates
Transcriptomics module Full-text querying
Clean data Result data
Raw data
Subjects Groups gene p-values
cell files Structured
expression z-values
querying
Events Protocols Profile-based analysis
Epigenetics module
Raw data Clean Resulting
Samples Assays Nimblegen CPG island Genome Study comparison
Illumina data Feature data
Web user interface
Faculty of Health, Medicine and Life Sciences

10. Generic Study Capture Framework
Data input / output
GSCF
Templates
Templates
Templates
Subjects Groups
xls, cvs, text
Data import
NCBO web
Events Protocols
Ontologies interface
Samples Assays
custom
custom
custom custom
custom
Molgenis programs
programs EBI custom
programs dbs
dbs
repository dbs

13. Used in European Projects
Food4me (Dublin)
NU-AGE (UNIBO, Bologna)
Bioclaims (UIB, Palma)
Nutritech (TNO, Zeist)
EuroDish (WUR, Wageningen)
ITFoM (proposed for metabolic syndrome studies)

14. Process the data…
Faculty of Health, Medicine and Life Sciences

15. Epigenetics DNA Methylation Pipeline
Raw data R
Nimblegen QC, processing Clean
DNA Result
Raw data R methylation data
Illumina QC, processing data Statistical with
(Genome analysis p-values
Feature (GFF)
Raw sequencing data Sequence Format)
MeDIP, BIS-Seq QC, processing

16. Connecting to Pathways:
1) Prepare data for pathway analysis
2) Connect processing pipelines
PathVisioRPC used from arrayanalysis.org
see: http://pathvisiorpc.wordpress.com
3) Store Pathway profiles as vectors,
Using pathways themselves as a vocabulary
C Evelo, K van Bochove & J Saito. Genes Nutr (2011) 6: 81-87Answering
biological questions - querying a systems biology database for
nutrigenomics
4) Allow queries for studies with same outcome
Faculty of Health, Medicine and Life Sciences

17. Integrate
Example
WikiPathway Pathway
Pathway on glycolysis.
Using modern systems
iology annotation.
And genes and
metabolites connected
to major databases.
Faculty of Health, Medicine and Life Sciences

18. Find the pathways:
Biological processes in duodenal mucosa affected by glutamine administration
number of genes
Pathway Changed Up Down Measured Total Z Score
Hs_Mitochondrial_fatty_acid_betaoxidation 6 6 0 16 16 4.456
Hs_Electron_Transport_Chain 17 17 0 85 105 4.278
Hs_Fatty_Acid_Synthesis 5 5 0 21 22 2.757
Hs_Fatty_Acid_Beta-Oxidation 6 6 0 31 32 2.424
Hs_mRNA_processing_Reactome 16 6 10 118 127 2.402
Hs_Unsaturated_Fatty_Acid_Beta_Oxidation 2 2 0 6 6 2.342
Hs_HSP70_and_Apoptosis 4 4 0 18 18 2.299
Hs_Oxidative_Stress 5 5 0 27 28 2.097
Hs_Fatty_Acid_Omega_Oxidation 3 3 0 14 15 1.915
Hs_Proteasome_Degradation 8 8 0 60 61 1.629
Hs_RNA_transcription_Reactome 5 5 0 38 40 1.25
Hs_Irinotecan_pathway_PharmGKB 2 1 1 12 12 1.154
Hs_Synthesis_and_Degradation_of_Ketone_Bodie
s_KEGG 1 1 0 5 5 1.023

19. Connecting to
other data
We both need
Study Capturing
Faculty of Health, Medicine and Life Sciences

20. If the mountain will not
come to Mahomet,
Mahomet must go to
the mountain.
Other repositories (like
dbXP!) have better
study descriptions.
Integrate in Sage
Synapse.
Pathway visualisation
missing: integrate
PathVisio in Synapse
(started).
Faculty of Health, Medicine and Life Sciences

21. PathVisio
www.pathvisio.org
• Data modeling and visualization on biological pathways
• Uses gene expression, proteomics and metabolomics data
• Can identify significantly changed processes
Martijn P van Iersel, Thomas Kelder, Alexander R Pico, Kristina Hanspers, Susan Coort, Bruce R Conklin, Chris
Evelo (2008) Presenting and exploring biological pathways with PathVisio. BMC Bioinformatics 9: 399

22. Understanding
genomics
Example
WikiPathways Pathway
Pathway on glycolysis.
Using modern systems
biology (MIM) annotation.
And genes and metabolites
connected to major
databases.
Faculty of Health, Medicine and Life Sciences

23. Faculty of Health, Medicine and Life Sciences

24. adding data =
adding colour
Example
PathVisio result
Showing proteomics
and transcriptomics
results on the glycolysis
pathway in mice liver
after starvation.
[Data from Kaatje
Lenaerts and Milka
Sokolovic, analysis by
Martijn van Iersel]
Faculty of Health, Medicine and Life Sciences

25. Download Pathways
Web services
SPARQL endpoint

26. How to do
data visualization?

27. Connect to Genome Databases

28. Backpages link to databases
Faculty of Health, Medicine and Life Sciences

29. BridgeDb
http://dx.doi.org/10.1186/1471-2105-11-5
Martijn van Iersel
BiGCaT Maastricht

30. Problem: Identifier Mapping
Entrez Gene
3643
?
Agilent probeset
A65_P12450

31. Solution: Built-in Mapping
• Generic
bioinformatics
platforms should
have identifier
mapping built-in.
BioConductor
PathVisio
Cytoscape
...
Batteries
Included

32. Problem: Which mapping service?
• Ensembl Biomart
• Synergizer
• CRONOS
• DAVID
• AliasServer
• MatchMiner
• OntoTranslate
or
• Local database

33. BridgeDB: Abstraction Layer
class
IDMapperRdb
relational database
interface
IDMapper class
IDMapperFile
tab-delimited text
class
IDMapperBiomart
web service
The BridgeDb Framework: Standardized Access to Gene, Protein and Metabolite Identifier
Mapping Services. Martijn P van Iersel, Alexander R Pico, Thomas Kelder, Jianjiong Gao, Isaac Ho,
Kristina Hanspers, Bruce R Conklin, Chris T Evelo. BMC Bioinformatics 2010, 11: 5.

34. CyThe- Network
saurus Merge Wiki
Tools PathVisio
Pathways
Cytoscape Plugins
BridgeDb
Internet webservices
Local Tab-
Mapping
BridgeDb Databas delimited
Services BioMart PICR - e text files
REST

35. BridgeDb interface
1: JAVA interface 2: REST interface

36. API Overview
BridgeDb.connect(...)
IDMapper.mapID(...)
Xref.getUrl()
DataSource.getUrl()

37. Easy & Flexible Code

38. REST API
http://webservice.bridgedb.org/Human/xrefs/L/1234
ILMN_1713029 Illumina
3255967 Affy
NP_001025186 RefSeq
IPI00005930 IPI
GO:0042752 GeneOntology
NM_033282 RefSeq
3255968 Affy
94233 Entrez Gene
ENSG00000122375Ensembl Human
234226_at Affy
A6NEB4 Uniprot/TrEMBL
0001780601 Illumina
GO:0008020 GeneOntology
606665 OMIM
A_23_P24234 Agilent
14449 HUGO

39. REST API
http://<Base URL>/<Species>/<function> [ /<argument> ... ]
http://webservice.bridgedb.org/Human/xrefs/L/1234
http://webservice.bridgedb.org/Human/search/ENSG00000122375
http://webservice.bridgedb.org/Human/attributeSet
http://webservice.bridgedb.org/Human/properties
http://webservice.bridgedb.org/Human/targetDataSources
http://webservice.bridgedb.org/Human/attributes/L/3643
http://localhost:8183/Human/xrefs/L/3643

40. R Example

41. Problem: Custom Microarrays
?
Custom probe
#QXZCY!34

42. Solution: Stacking
EnsMart
Custom
table

43. CyThesaurus

44. MIRIAM and Identifiers.org
Regular
expression for
autodetection Pattern for
generating URLs
Link to
documentation

45. Availibility
BMC Bioinformatics. 2010 Jan 4;11(1):5.
www.bridgedb.org
www.helixsoft.nl/blog bridgedb-discuss@googlegroups.com

46. Innovate using BridgeDB
Data
Metabolite
Flux
Visualizing fluxes on metabolic pathways 46

47. Integrating it all
Visualizing fluxes, data and annotation

48. Extending pathways, how to do it?
Faculty of Health, Medicine and Life Sciences

49. Network approaches to extend pathways
E.g. most pathways don’t have miRNA’s

50. Adding miRNA’s

51. Pathway Loom, weaving pathways
Faculty of Health, Medicine and Life Sciences

52. Faculty of Health, Medicine and Life Sciences

53. Adding miRNA’s clutters

54. PathVisio RI plugin provides backpage info
microRNAs in pathway analysis. The Regulatory Interaction plugin offers a suitable middle-ground between not including any
miRNAs in pathways, which misses this regulatory information, and including all validated miRNA-target interactions, which
clutters the pathway. After loading interaction file(s), selecting a pathway element shows the interaction partners of this
element and their expressions in a side panel. This allows for the detection of potential active regulatory mechanisms in the
study at hand.
http://www.bigcat.unimaas.nl/wiki/images/f/f6/VanHelden-poster-nbic2012.pdf

55. Or consider pathway as a network
Faculty of Health, Medicine and Life Sciences

56. GPML Cytoscape Plugin
http://www.pathvisio.org/wiki/Cytoscape_plugin

57. Cytoscape visualization used to group
PPS1
Liver
All pathways
Pathways with high z-score
grouped together.
Explains why there are
relatively few significant
genes, but many pathways
with high z-score.
Robert Caesar et al (2010) A combined transcriptomics and lipidomics analysis of subcutaneous,
epididymal and mesenteric adipose tissue reveals marked functional differences. PLoS One 5: 7. e11525
http://dx.doi.org/doi:10.1371/journal.pone.0011525

58. Explore pathway interactions
Thomas Kelder, Lars Eijssen, Robert Kleemann, Marjan van Erk, Teake Kooistra, Chris Evelo
(2011) Exploring pathway interactions in insulin resistant mouse liver BMC Systems Biology 5: 127
Aug. http://dx.doi.org/doi:10.1186/1752-0509-5-127

59. What we used
Non-redundant shortest paths in a weighted
graph.
1. A set of pathways
2. An interaction network
3. Weight value for all edges
= experimental expression of connected
genes.

60. Pathway interactions and what causes them

61. An indirect interaction between the Axon Guidance and Insulin Signaling pathways in the network for
the comparison between HF and LF diet at t = 0. Left: Network representation of the identified path
between the two pathways, consisting of three proteins Gsk3b, Sgk3 and Tsc1. Right: The location of these
proteins in the KEGG pathway diagrams. The newly found indirect interactions have been added in red.

62. Pathway interactions and
detailed network visualization
for the interactions with three
apoptosis related pathways for
the comparison between HF and
LF diet at t = 0. A: Subgraph of the
pathway interaction network, based
on incoming interactions to three
stress response and apoptosis
pathways with the highest in-
degree. Pathway nodes with a thick
border are significantly enriched (p
< 0.05) with differentially expressed
genes. B: The protein interactions
that compose the interactions
between the three apoptosis
related pathways and their
neighbors in the subgraph as
shown in box A (see inset, included
interactions are colored orange).
Protein nodes have a thick border
when their encoding genes are
significantly differentially expressed
(q < 0.05).

63. We tried to make it easier with
The CyTargetLinker Cytoscape Plugin
Extending pathways on the fly.
Provided databases with the plugin:
• miRNAs with targets
• Transciption Factors with targets
• Drug – Target Interactions
• ENCODE derived databases
Extend with your own.

64. MiRNAs of Interest
miRNA target information from mirTarBase

65. miRTarBase as a target interaction network
Collection of miRNA-target gene interactions in the miRTarBase database with 1,715 genes,
286 miRNAs and 2,817 interactions.

66. miRNAs associated with colorectal cancer
extended with validated target genes

67. human ErbB signaling pathway extended
with validated microRNA regulation

70. OPS Framework
OPS GUI Architecture. Dec 2011
App
Framework
Web Service API Sparql Web
Services
OPS Data Model
Identity &
Vocabulary
Management Semantic Data Workflow Engine
RDF Data Cache
Chemistry
Normalisation &
Registration Descriptor Descriptor
Descriptor Descriptor Nanopub Nanopub
Feed in WikiPathways
RDF 1
relationships, use BioPAX RDF 2 RDF 3 RDF 4
to create the RDF
Public
Vocabularies Data 1 Data 2 Data 3 Data 4

71. And then we have linked data?

72. Well yes, for Open PHACTS we do…
OPS Data Model
Identity &
Vocabulary
Management Semantic Data Workflow Engine
Chemistry
Normalisation &
Registration
Descriptor Descriptor
RDF 1 RDF 2
Public
Vocabularies Data 1 Data 2

73. But really…,
what about federated SPARQL queries?
Descriptor Descriptor
RDF 1 RDF 2
Other
Public
Vocabularies Data 1 Data 2 Public
Vocabularies

74. Most often partly…
If the vocabularies used are different linking just database IDs not good enough.
We need full mappings of ontologies.
Identification of overlapping modules.
And maybe… Suggestions for ontologies to use in specific field.
Identity
Mapping
Descriptor Descriptor
RDF 1 RDF 2
Other
Public
Vocabularies Data 1 Data 2 Public
Vocabularies

75. Thanks!
WikiPathways team:
• Martijn van Iersel (PathVisio,
BridgeDB)
• Thomas Kelder (WikiPathways,
networks)
• Alex Pico (US team leader)
• Brice Conklin (former US team leader)
• Kristina Hanspers (US curation)
• Martina Kutmon (CyTargetLinker)
• Susan Coort (Regulatory plugins)
• Lars Eijssen (Data pipelines)
• Anwesha Dutta (Flux visualisation)
• Andra Waagmeester (LOOM)
• Egon Willighagen (Open Phacts)
Funding. Dutch: IOP, NBIC, NuGO, NCSB. Regional:
Transnational University. EU: NuGO and Microgennet,
IMI: Open Phacts + Agilent thought leader grant and
NIH.

76. Thanks!
Funding. Dutch: IOP, NBIC, NuGO, NCSB. Regional:
Transnational University. EU: NuGO and Microgennet,
IMI: Open Phacts + Agilent thought leader grant.

77. Analyzing GO representation in
pathways using an independent
library for ontology analysis
Combining efforts and information to
increase biological understanding

78. Structuring biological data
• Gene Ontology (GO)
– Protein function or
localization
– Hierarchically structured
terms
– 3 topics (namespaces)
• Biological process
• Molecular function
• Cellular component
– Disadvantage
• No information on interactions

79. Structuring biological data
• Pathways
– Network of interactions
– Structural overview of elements in the
pathway
– Disadvantages:
• Missing structure
of interacting
pathways
• Overlap and
abundance in
pathways

80. Analysis based on structures
• Uses:
– Better overview of the data
– Increased biological understanding
• Challenges in the field:
– Difficulty comparing algorithms
– Good work may be overlooked
– Redundant efforts
– Out-of-date algorithms used
– Comparison extremely difficult

81. Goals:
• Develop an independent library for ontology
analysis in which efforts can be combined
• Increase biological understanding by
combining knowledge on pathways and gene
ontology.

82. Independent library for ontology
analysis
• Open source:
– Collaboration
– Clear view of the algorithm
– Free use
– Minimalizing redundant efforts
• Usable for multiple ontology's and identifiers

83. Combining Pathways and GO
• Display information on the function of the
pathway
• Make a comparison between pathways
• Quality control
– Single pathway
– List of pathways

84. Materials
• PathVisio
– Open source Tool for visualizing and analyzing
pathway data
• BridgeDb
– id mapping framework for bioinformatics
• WikiPathways
– Community curated pathway data source

85. Independent Library
• Manager input:
1. Ontology Terms
(File)
2. Map of term with
identifier
3. Method Selection

86. Methods
Id’s linked Genes not
to GO linked to GO
Id’s in
pathway a b a+b
Id’s not in
pathway c d c+d
a+c b+d n

87. Plug-in
• Panel for the analysis of a single pathway
– Display GO terms in a table with score
– Highlight matches
– Save results
• Menu Item for analyzing a list of pathways
– Select a folder containing pathway files
– Individual result files
– File containing all results with extra info

88. Single Pathway analysis

89. Single Pathway analysis
• Regulation of blood pressure
• Angiogenesis
• Others:
– G-protein coupled receptor
– proteolysis
Homo sapiens: Mus musculus:
name score name score
G-protein coupled receptor signaling kidney development 50%
pathway 35% G-protein coupled receptor signaling
regulation of cell proliferation 29% pathway 50%
proteolysis 29% response to drug 37%
regulation of blood pressure 29% negative regulation of cell proliferation 37%
response to drug 29% positive regulation of apoptotic process 37%
regulation of vasoconstriction 29% regulation of blood pressure 37%
positive regulation of apoptotic process 29% response to salt stress 25%
negative regulation of cell growth 23% regulation of systemic arterial blood
kidney development 23% pressure by circulatory renin-angiotensin 25%
elevation of cytosolic calcium ion arachidonic acid secretion 25%
concentration 23% blood vessel development 25%

90. Multiple Pathway analysis

91. Multiple Pathway analysis
0 2 4 6 8 10 12 14 16 18
Biological Process
12 of 105 terms signal transduction
xenobiotic metabolic process
oxidation-reduction process
metabolic process
G-protein coupled receptor signaling pathway
gene expression
nerve growth factor receptor signaling pathway
apoptotic process
synaptic transmission
DNA repair
mitotic cell cycle
innate immune response
0 10 20 30 40 50 60 70 80
Cellular Compontent
cytoplasm
12 of 26 terms cytosol
nucleus
plasma membrane
membrane
integral to membrane
mitochondrion
nucleoplasm
endoplasmic reticulum membrane
extracellular region
endoplasmic reticulum
integral to plasma membrane
microsome
extracellular space

92. Goals:
• Develop an independent library for ontology
analysis in which efforts can be combined
• Increase biological understanding by
combining knowledge on pathways and gene
ontology.

93. Independent library
• Reads GO terms from file
• Mapping from term to identifier
• Analysis on sample data
• Framework enables more methods to be
added

94. Combining Pathways and GO
• Single Pathway:
– More information on pathway
– Quality control possible
• Pathway List:
– Separate results for every pathway
– Enables structuring possibility’s
– Quality control possible