1. Strategies & Examples for Functional Modeling
COST Functional Modeling Workshop
22-24 April, Helsinki

2. Types of data sets and modeling
Commercial array data – more likely to have tools that support the use of array IDs.
Custom/USDA array data – problems with updating IDs, linking to function and using array IDs directly in functional modeling tools.
Proteomics data – larger data sets; need to make background references to determine enrichment.
RNA-Seq data – largerand more complex data sets; novel transcripts currently can’t be included in modeling (contact AgBase to assign GO).
Real-time data or quantitative proteomics data – hypothesis testing.

3. Functional Modeling Strategies
GO summary (using Slim sets)
GO enrichment (statistical!)
Pathways analysis
Interaction or networks analysis
Hypothesis testing
Note:
Functional modeling should be integrated.
Approaches are complementary, not exclusive.
Modeling is driven by the biology (not the other way round).

4. Modeling Strategy Think about using multiple functional approaches.
GO, pathways, networks
complementary
What is available for your species?
What GO is available?
What species does the pathways/network analysis use?
What resources do you have?
at your institute (e.g. commercial pathways analysis)
open source (e.g. GO Enrichment analysis)
using online vs installed
Iterative – further functional modeling based on initial results
GO hypothesis testing?

5. 1. GO Functional Summary
high throughput data sets gives us 1000s -10,000s of gene products
can’t know everything about all gene products
tendency to ‘cherry pick’ ones you recognize
instead, can group gene products by function
this gives us a manageable number of categories to process
enables us to see trends, patterns, etc
Use GO Slim sets to ‘summarize’ data
Lose details (but can gain perspective).
Some GO Slim sets are ageing – not being updated as changes to the GO are made.
Different Slim sets have different terms – which is best for your data?
AgBase GOSlimViewer tool.

6.
The Slim set you use matters - need to determine which one to use & report it in Methods.

9. Functional Summary
Not all GO terms are annotated equally, e.g., metabolism!
can slim the complete GO for a species as a background set and then determine terms in your data are disproportionately expressed.
Can use Slims to compare two data sets (e.g., control vs treatment).
Use Slims for your own sanity – are you seeing what you expect to see?

10. Membrane proteins grouped by GO BP:
B-cells
Stroma
cell cycle/cell proliferation
cell adhesion
cell growth
apoptosis
immune response
ion/proton transport
cell migration
cell-cell signaling
function unknown
development
endocytosis
proteolysis and peptidolysis
signal transduction
protein modification

11. Membrane proteins grouped by GO BP:
B-cells
Stroma
cell cycle/cell proliferation
apoptosis
immune response
cell migration
cell-cell signaling
function unknown

12. BVDV Infection – cytopathic (CP) vs non-cytopathic (NCP) infection
(comparing function between 2 different conditions)

14. 2. Determining over-represented or under-represented function.
most typically used functional analysis method
many, many tools that do this – see:
very different visualization
will use some of these tools in practical session

16. Some useful expression analysis tools:
Database for Annotation, Visualization and Integrated Discovery (DAVID)
AgriGO -- GO Analysis Toolkit and Database for Agricultural Community
used to be EasyGO
chicken, cow, pig, mouse, cereals, dicots
adding new species by request
Onto-Express
can provide your own gene association file
Ontologizer
WebStart widget (requires Java); now on Galaxy
requires OBO file & GAF (enables users to select their own annotations)

17. GO Enrichment tools that support agricultural species.

20. structurally and functionally re-annotated a microarray
quantified the impact of this re-annotation based on GO annotations & pathways represented on the array
tested using a previously published experiment that used this microarray
re-annotation allows more comprehensive GO based modeling and improves pathway coverage
re-annotation resulted in a different model from previously published research findings

22. Evaluating GO tools Some criteria for evaluating GO Tools:
Does it include my species of interest (or do I have to “humanize” my list)?
What does it require to set up (computer usage/online)
What was the source for the GO (primary or secondary) and when was it last updated?
Does it report the GO evidence codes (and is IEA included)?
Does it report which of my gene products has no GO?
Does it report both over/under represented GO groups and how does it evaluate this?
Does it allow me to add my own GO annotations?
Does it represent my results in a way that facilitates discovery?

23. RNASeq GO Enrichment
RNASeq experiments: longer transcripts and more highly expressed transcript are more likely to be differentially expressed.
Current GO enrichment tools do not account for RNASeq platform bias (most based upon arrays).
assume that all genes are independent and equally likely to be selected as DE

24. 3. Pathway Analysis Freely available tools:
from public databases, e.g. KEGG & Reactome
Freely available tools, e.g. Cytoscape
Commercial pathways analysis tools: e.g., Ingenuity Pathways Analysis (IPA), Pathway Studio, etc.
some tools only have limited species – need to “humanize” animal data, etc for plants with Arabidopsis
everything gives you cancer
Many pathways analysis tools combine pathways analysis, network analysis.

25. Reactome Skypainter

26. KEGG Pathways

27. Analysis tools (commercial)
Networks
Ingenuity Pathway
Analysis
Pathways
functions and diseases
Gene Ontology (GO) groups
Pathway Studio
GSEA
Pathways
IPA analysis included as IPA.txt

28. Data Curation
Ingenuity: Manually curated database by Ph.D level scientists (mining 32 different peer reviewed journals).
Pathway studio: Automated curation by Medscan Reader using Natural language processing (NLP) technology. Mining Pubmed abstracts and peer reviewed journals
users can do their own text mining

29. (Comparison by Divya Peddinti)
Comparison Criteria
Features
Proportion of proteins involved in modeling
Data generation
Display
Test Dataset: 3,600 bovine spermatozoa proteins
(Comparison by Divya Peddinti)

30. Feature
Ingenuity Pathway analysis (IPA)
Pathway studio
Input
GI number
Microarray ID
Affymetrix ID
GenBank
Swiss Prot Accession
Unigene ID
Name orAlias
HUGO ID
Entrez gene
Name or Alias
HUGO ID
Databases
Contains biological interactions data for human, mouse, rat Orthologous mapping available for dog, Cow, Chimp, Chicken, Rhesus macaque monkey, Arabidopsis thaliana, Saccharomyces cerevisiae, Drosophila melanogaster, Caenorhabditis elegans, Danio rerio
Contains biological data for human, mouse, rat, bacteria, chicken, Zebra fish, frog, cow, bee, dog, Arabidopsis, Drosophila, Yeast, and transplantation research etc.

31. Builds networks with a maximum of 35 genes/ proteins
Ingenuity Pathway analysis (IPA)
Pathway studio
Statistical test
The significance value (p value) assigned to the function / pathways using Fischer’s exact test
The statistical significance of the overlap between the protein list and a GO group or pathway using the Fischer’s exact test.
Updates
Quarterly
Networks
Builds networks with a maximum of 35 genes/ proteins -

32. Proteins involved in modeling

33. Data generation 37 7 26

34. Pathway display EGF signaling pathway

35. 4. Network Analysis
IPA & Pathway Studio equally efficient at drawing networks of relationships.
IPA : simplifies the pathway display and creates more manageable user friendly network for users to analyze.
Pathway Studio: Shows the relations in a table format.
STRING Database - known and predicted protein interactions.

38. 5. Hypothesis Testing
high throughput data sets – ‘fishing expedition’ or hypothesis generation
but GO also serves as a repository of biological function – can be used for hypothesis testing based on these data sets

39. The critical time point in MD lymphomagenesis18 16
Genotype
Hypothesis
At the critical time point of 21 dpi, MD-resistant genotypes have a T-helper (Th)-1 microenvironment (consistent with CTL activity), but MD-susceptible genotypes have a T-reg or Th-2 microenvironment (antagonistic to CTL). 14
Susceptible (L72)
Resistant (L61) 12
mean total lesion score 10
Non-MHC associated resistance and susceptibility 8 6 4 2 20 40 60 80
100
days post infection 39

40. CYTOKINES AND T HELPER CELL DIFFERENTIATION
T reg
NAIVE CD4+ T CELL
APC
Th-2
Th-1
Shyamesh Kumar

41. Th-1, Th-2, T-reg ? Inflammatory? T reg IL 12 IL 4 NAIVE CD4+ T CELL
L6 Whole
APC
L7 Whole
Smad 7
L7 Micro
IL 12
IL 4
Th-1, Th-2, T-reg ?
Inflammatory?
Th-2
Th-1
TGFβ
IL 4
IL10
IFN γ
IL 12
IL 18
CTL
Macrophage
NK Cell 41

42. Step II. Multiply by quantitative data for each gene product.
Step III. Inclusion of quantitative data to the phenotype scoring table and calculation of net affect.
Step I. GO-based Phenotype Scoring. 1 -1
SMAD-7
GPR-83
CTLA-4
TGF-b
IFN-g
IL-18 ND
IL-13
IL-12
IL-10
IL-8
IL-6
IL-4
IL-2
Inflammation
Treg
Th2
Th1
Gene product
ND = No data
Gene product
Th1
Th2
Treg
Inflammation
IL-2
1.58
-1.58
IL-4
0.00
IL-6
-1.20
1.20
IL-8
1.18
IL-10
IL-12
IL-13
1.51
-1.51
IL-18
0.91
IFN-g
TGF-b
-1.71
1.71
CTLA-4
-1.89
1.89
GPR-83
-1.69
1.69
SMAD-7
Net Effect
-1.29
-5.38
10.15
-5.98
Step II. Multiply by quantitative data for each gene product.

43. Microscopic lesions L6 (R) L7 (S) 60 50 40 Net Effect 30 20 10 Th-1
5mm
Microscopic lesions 60
L6 (R) 50 40
L7 (S)
Net Effect 30 20 10
Th-1
Th-2
T-reg - 10
Inflammation
Phenotype - 20

44. L6 Resistant L7 Susceptible Pro T-reg Pro T-reg Pro Th-1 Pro Th-2 Anti
Anti CTL
Pro CTL
Anti CTL
Pro CTL

46. Concluding thoughts on functional modeling.
“By doing just a little every day, I can gradually let the task overwhelm me.”
Ashleigh Brilliant

47. Bringing it all together…
There is no one “correct” way; there is no “right” answer.
Using multiple functional modeling strategies (e.g., GO, pathways, networks) can help with insights.
Need to use biological knowledge to bring these different approaches together.
Functional modeling is often iterative.
Need to focus not only on what is known but what is new!

48. Overview of Functional Modeling Strategy
Genes/Proteins with no GO annotations
Microarrays
ArrayIDer
GORetriever
GOanna
Blast2GO
Protein/Gene identifiers
Proteomics
GO annotations
Genome2seq
RNASeq
GO Enrichment analysis
Ingenuity Pathways Analysis (IPA)
Pathway Studio
Cytoscape
DAVID
AgriGO
Onto-tools
GOSlimViewer
AutoSlim
Pathways and network analysis
Ingenuity Pathways Analysis (IPA)
Pathway Studio
Cytoscape
DAVID
Yellow boxes represent AgBase tools
Green boxes are non-AgBase resources

49. Functional Modeling Considerations
Should I add my own GO?
use GOProfiler to see how much GO is available for your species
use GORetriever to find existing GO for your dataset
Does analysis tool allow me to add my own GO?
Should I do GO analysis and pathway analysis and network analysis?
different functional modeling methods show different aspects about your data (complementary)
is this type of data available for your species (or a close ortholog)?
What tools should I use?
which tools have data for your species of interest?
what type of accessions are accepted?
availability (commercial and freely available)

50. Some Limitations Annotation is not complete.
not all the data is annotated
some gene products have no functional information
Gene Ontology is only one aspect of functional modeling.
anatomy, tissue expression, phenotype, disease, etc
Gene nomenclature – need to know what we are annotating!
Functional modeling tools need to handle larger data sets (& multiple ontologies?).