1. Extracting Biological Information from Gene Lists
Simon Andrews, Laura Biggins, Boo Virk
v1.0

2. Analysis doesn’t end here!
Analysis of processed sample: Data acquisition – sequencing, microarray analysis, mass spectrometry
Biological material
Sample for analysis
Isolation of DNA, RNA or proteins
Sample processing
Analysis doesn’t end here!
Raw data file(s)
Results Table
Containing hits – genes, transcripts or proteins
Public databases
Data analysis: identification of genes, transcripts or proteins

3. Why functional analysis?
Advantages:
Biological insight
Validation of experiment
Generate new hypothesis
Limitations:
Amount of information depends on the species
Will only find known/published links between genes
If working on something novel – information available may be limited

4. What this course covers
Morning
Introduction to Gene Lists
Gene List Practical
Coffee
Presenting results
Presenting Results Practical
Afternoon
Motif Searching
Motif Searching Practical
Coffee
Networks and Interactions
Network Practical
Commercial tools

5. Gene Lists Types of gene list: Names of genes
Names of genes ordered by qualitative value
Names of genes ordered by quantitative value
Gene lists can be ranked
P-value
Other Stat
Ordered
Gene lists can be filtered
Cut off point
Subset of genes

6. Transforming Gene ID’s
Need to use relevant ID to extract information from databases
BioMart ID conversion tool allows us to do this easily and quickly online

9. Download this data, import transformed ID’s into table

10. I have my gene list, what next?
Hyperlinked table:
Gene
UniProt
Name
Score
Reactome
TP53
P04637
Tumor Suppressor p53
125527
CDK1
P06493
Cyclin-dependent kinase 1
113740
POLE
Q07864
DNA Polymerase Epsilon
107190
KPNB1
Q14974
Importin subunit beta-1
35542
CHEK1
O14757
Serine/threonine-protein kinase Chk1
35271
AURKB
Q96GD4
Aurora kinase B
30803
RPA2
P15927
Replication protein A 32 kDa subunit
22207
CDT1
Q9H211
DNA replication factor Cdt1
21735
MCMBP
Q9BTE3
MCM complex-binding protein
17811
TUBG1
P23258
Tubulin gamma-1 chain
16895
RAN
P62826
GTP-binding nuclear protein Ran
16384
RANGRF
Q9HD47
Ran guanine nucleotide factor Mog1
15527
BLM
P54132
Bloom syndrome protein
14883
PCNA
P12004
Proliferating Cell Nuclear Antigen
13982
SETD8
Q9NQR1
Pr-Set7
13711
RCC1
P18754
Regulator of chromosome condensation
13302
MCM5
P33992
DNA replication licensing factor MCM5
12806
CDC25C
P30307
M-phase inducer phosphatase 3
12510
PLK1
P53350
Serine/threonine-protein kinase PLK1
10930
MZT1
Q08AG7
Mitotic-spindle organizing protein 1
9210

11. Hyperlinked tables Advantages However…
Easy to create –no special data-mining software needed
One-click direct access to relevant pages
Reference resource
However…
Need to become familiar with the resources available - tailor hyperlinks to be specific for your organism and questions being asked
Information on one gene at a time in your gene set
Need to get relevant resource ID’s

12. I have my gene list, what next?
Annotated table:
Gene
UniProt
Name
Score
PANTHER GO-Slim BP
CHEK1
O14757
Serine/threonine-protein kinase Chk1
35271
apoptotic process;nitrogen compound metabolic process;biosynthetic process;transcription from RNA polymerase II promoter;cellular protein modification process;cell cycle;cell communication;apoptotic process;response to stress;response to abiotic stimulus;regulation of transcription from RNA polymerase II promoter;regulation of cell cycle;chromatin organization
MCM5
P33992
DNA replication licensing factor MCM5
12806
cell cycle;cell communication
RCC1
P18754
Regulator of chromosome condensation
13302
cellular component movement;mitosis;chromosome segregation;cellular component morphogenesis;intracellular protein transport;cellular component organization
CDC25C
P30307
M-phase inducer phosphatase 3
12510
DNA replication;cell cycle
PLK1
P53350
Serine/threonine-protein kinase PLK1
10930
DNA replication;DNA repair;DNA recombination;cell cycle
TP53
P04637
Tumor Suppressor p53
125527
glycogen metabolic process;protein phosphorylation;mitosis;cell communication
CDK1
P06493
Cyclin-dependent kinase 1
113740
nitrogen compound metabolic process;biosynthetic process;DNA replication;RNA metabolic process;cellular process;regulation of biological process;regulation of catalytic activity
BLM
P54132
Bloom syndrome protein
14883
nucleobase-containing compound metabolic process;cell cycle;cell communication;RNA localization;intracellular protein transport;nuclear transport
RPA2
P15927
Replication protein A 32 kDa subunit
22207
nucleobase-containing compound metabolic process;mitosis;nucleobase-containing compound transport;regulation of catalytic activity
TUBG1
P23258
Tubulin gamma-1 chain
16895
phosphate-containing compound metabolic process;cellular protein modification process;cell cycle
KPNB1
Q14974
Importin subunit beta-1
35542
phosphate-containing compound metabolic process;protein phosphorylation;cytokinesis;cell cycle;regulation of cell cycle;chromatin organization;cytoskeleton organization
MZT1
Q08AG7
Mitotic-spindle organizing protein 1
9210
protein targeting;nuclear transport
PCNA
P12004
Proliferating Cell Nuclear Antigen
13982
RAN
P62826
GTP-binding nuclear protein Ran
16384
POLE
Q07864
DNA Polymerase Epsilon
107190
AURKB
Q96GD4
Aurora kinase B
30803
MCMBP
Q9BTE3
MCM complex-binding protein
17811
CDT1
Q9H211
DNA replication factor Cdt1
21735
RANGRF
Q9HD47
Ran guanine nucleotide factor Mog1
15527
SETD8
Q9NQR1
Pr-Set7
13711

13. Annotation Sources
There are many databases for annotation sources, including (but not limited to):
Gene Ontology (GO)(most popular)
Pathways and Interactions
Protein domains
Co-expression
Transcription binding sites

14. Annotation Sources
There are many databases for annotation sources, including (but not limited to):
Gene Ontology (GO)(most popular)
Pathways and Interactions
Protein domains
Co-expression
Transcription binding sites

15. What is Gene Ontology (GO)?
Collaborative effort addressing need for consistent descriptions of gene products across different databases
GO project has three structured ontologies describing gene products independent of species:
Biological Processes (BP),
Cellular Components (CC)
Molecular Functions (MF)

16. GO Structure 3 GO domains: Root ontology terms general Parent specific1 2 3
Root ontology terms
general
specific
Parent
Child

17. Subsets of GO terms GO slim terms: GO fat terms:
Cut-down versions of the GO ontologies that contain a subset of terms from the GO resource
Give a broad overview of the ontology content without the detail of the specific, fine-grained terms
GO fat terms:
subset comprising more specific terms

18. Annotation Sources
There are many databases for annotation sources, including (but not limited to):
Gene Ontology (GO)(most popular)
Pathways and Interactions
Protein domains
Co-expression
Transcription binding sites

19. Pathway and Interactions
Are specific pathways enriched in my list?
What other genes are in this pathway?
Which genes/gene products interact with my genes of interest?
Databases include:

20. Annotation Sources
There are many databases for annotation sources, including (but not limited to):
Gene Ontology (GO)(most popular)
Pathways and Interactions
Protein domains
Co-expression
Transcription binding sites

21. Protein Domain Can I find shared protein domains?
What is the function of shared domain?
Which other proteins share this domain?
Databases include:

22. Annotation Sources
There are many databases for annotation sources, including (but not limited to):
Gene Ontology (GO)(most popular)
Pathways and Interactions
Protein domains
Co-expression
Transcription binding sites

23. Co-expression Which genes are co-expressed?
Automatic grouping of genes (rather than human curation (GO))
Databases include:

24. Annotation Sources
There are many databases for annotation sources, including (but not limited to):
Gene Ontology (GO)(most popular)
Pathways and Interactions
Protein domains
Co-expression
Transcription binding sites

25. Annotated tables Advantages
Information on function from larger gene sets
Sort groups of genes (GO term, pathway, protein domain)
Relatively easy to create
Reference resource
However…
Need to become familiar with the resources specific to your research
Lots of information can be difficult to sort efficiently

26. What does functional information tell me?
Have functional information about a gene set
Can verify genes implicated in experiment are functionally relevant, and to discover unexpected shared functions
Determine which functions are enriched in gene set
How?
Compare to a background list of genes

27. What is a background list?
In theory, any gene that could have been differentially expressed in your experiment
RNA seq – all genes apart from those with less than reads
Arrays – all genes in the array
ChipSeq - any gene on the chip. vs

28. Choosing a Background List
Which background list to use?
Whole set of genes
Tissue/cell specific genes
Manually made list, derived from your experiment and analysis vs ?
Choice of background list place where things are likely to go wrong. Experimental design and making the most appropriate comparison
Manual list e.g. based on expression level

29. Statistics to test for enrichment
13,101 genes on chip
Related to disease
260/747 = 34.8%
Gene List
Are these proportions the same?
3005 genes related to disease
3005/13,101=
23.1%
Do not related to disease
487/747 = 65.2%

30. Lots of Statistical tests to choose from
• Hypergeometric test • Fisher’s exact/Chi-squared • Binomial • Kolmogorov Smirnov • Permutation

31. Hypergeometric test
Uses hypergeometric distribution to measure the probability of having drawn a specific number of successes (out of a total number of draws) from a population
Example:
Imagine that there are 4 green and 16 red marbles in a box.
You close your eyes and draw 5 marbles without replacement
What is the probability that exactly 2 of the 5 are green?

32. Are these proportions the same?
Gene List
3005 genes map to disease
3005/13,101=
23.1%
13,101 genes on chip
Map to disease
260/747 = 34.8%
Do not map to disease
487/747 = 65.2%
Are these proportions the same?
What is the probability (p-value) that exactly 260 genes (out of 747) map to disease, given that there are 3005 of those genes in the background (13,101 genes)?

33. Hypergeometric test Limitations: Assumes independence of categories
Input
Sample Size
Output
Specifics
Hypergeometric
Unranked/Ranked
List
Large (5% of background)
P-value
Finite population – probability of success changes
Limitations:
Assumes independence of categories
Result terms often include directly related terms
Is there really evidence for both terms?
Works better with larger samples (5% of background)

34. Based on Hypergeometric test:
Input
Sample Size
Output
Specifics
Hypergeometric
Unranked/Ranked
List
Large (5% of background)
P-value
Finite population – probability of success changes
Based on Hypergeometric test:
Test
Input
Sample Size
Output
Specifics
Fisher’s Exact
Unranked/Ranked
List
Small
P-value
Can be used to compare 2 conditions as well as gene list to background
one-tailed or two-tailed
Binomial
Large
Does not assume finite population – probability of success remains the same

35. Limitations of Fisher’s Exact and Binomial test
Neither account for variation in the number of genes annotated to individual terms/functions being tested
or the number of terms/functions associated with individual genes
Therefore, tend to over-estimate significance if the gene set has an unusually high number of annotations
Assume independence of categories

36. Lots of Statistical tests to choose from
• Hypergeometric
• Fisher’s exact/Chi-squared
• Binomial
• Kolmogorov Smirnov
• Permutation
Used for ranked gene lists only
Output: enrichment scores (ES) for functions, which can then be translated into a p-value

37. Multiple testing correction
Error types in statistics:
Statistical
Decision:
True state in Gene List
Not Overrepresented
Overrepresented
Significant
Type I error
(False Positive)
Correct
Not Significant
Type II error
(False Negative)
Traditionally, a test or a difference are said to be “significant” if the probability of type I error is: α =< 0.05

38. Probability of error increases from 5% to 14.3%
Example: You want to compare 3 groups and you carry out 3 hypergeometric tests, each with a 5% level of significance (P<0.05)
Probability of not making type I error = 95% = (1 – 0.05)
Overall probability of no type I errors is:
0.95 * 0.95 * 0.95 = 0.857
Therefore probability of at least one type I error is:
= or 14.3%
If comparing 5 groups instead of 3, the multiple testing error rate is 40%! (=1-(0.95)n)
Solution for multiple comparisons:
Multiple testing correction
Probability of error increases from 5% to 14.3%

39. Multiple test corrections
Bonferroni
Significant level (e.g. 0.05) /number of tests = new threshold
This is an over correction if tests are correlated
Benjamini-Hochberg
Rank the p-values
Apply more stringent correction to the most significant, and least stringent to the least significant p-values

40. Statistical issues
• We want to Identify functions of maximal biological significance – BUT this is not perfectly correlated with statistical significance • Use p‐values as a tool to rank functions but don’t take them too literally • Need to correct for multiple testing

41. Tools for functional gene list analysis
There are many different tools available, both free and commercial
Popular web-based tools include:

42. PANTHER (Protein ANnotation THrough Evolutionary Relationship) http://www.pantherdb.org/
One of the most widely used online resources for gene function classification and genome wide data analysis
PANTHER users have successfully analysed data from:
Gene expression
Proteomics
Genome-wide association study (GWAS) experiments
PANTHER is part of the GO consortium, thus PANTHER annotation = up to date GO curation

43. PANTHER for functional classification

48. Send list to > File
Saves table in a tab delimited .txt file

49. PANTHER for statistics

51. Annotations from PANTHER include:
GO-slim terms
PANTHER “protein class”
PANTHER “Pathway” terms
Doesn’t cluster together genes with similar GO terms in table
Statistics: Binomial test with Bonferroni multiple testing correction

52. https://david.ncifcrf.gov/
Gathers data from many different databases – this is customisable
Functional Clustering
Uses many annotations, including GO-Fat terms – more specific set of GO terms
Statistics: Fisher’s Exact Test and multiple testing correction

53. DAVID for functional classification

62. Functional Clustering
Enrichment score for the whole cluster rather than individual functions, DAVID anything above 2 or 3 is considered as enriched

63. Which DAVID tool should I use?

64. GOrilla http://cbl-gorilla.cs.technion.ac.il/

65. Which tool to use?
Choose a tool that: – Includes your gene / probe identifiers – Includes your species – Has up‐to‐date annotation – Lets you define your background (if possible) – Try a few different tools – Try gene lists of varying length