This document summarizes a presentation about automatic annotation in UniProtKB. It discusses how UniProtKB uses automatic annotation systems like UniRule and SAAS to annotate the increasing number of sequences as manual annotation cannot keep pace. UniRule allows manual creation of annotation rules while SAAS uses machine learning. It also describes how UniProtKB defines and works to annotate complete proteomes from various organisms.
2. Talk outline
• Introduction to UniProt
• UniProtKB annotation and propagation
• Data increase and the need for Automatic Annotation
• Automatic annotation systems in UniProtKB
• UniRule Automatic Annotation System
• Complete Proteomes in UniProtKB
30 January 2015
2
3. Talk outline
• Introduction to UniProt
• UniProtKB annotation and propagation
• Data increase and the need for Automatic Annotation
• Automatic annotation systems in UniProtKB
• UniRule Automatic Annotation System
• Complete proteomes in UniProtKB
30 January 2015
3
4. 30 January 2015
4
UniProt Consortium
• Formed in 2002
• Previously known as “Swiss-Prot” since 1986
• UniProt group at the EBI is led by Claire Odonovan and Maria
Jesus Martin, part of the PANDA proteins group led by Rolf
Apweiler
• UniProt group at PIR, Georgetown University is led by Cathy Wu
• UniProt group at SIB (Geneva/Lausanne) is led by Ioannis
Xenarios and Lydie Bougeleret (heirs to Amos Bairoch, left 2009)
• UniProtKB is UniProt KnowledgeBase, and includes TrEMBL and
Swiss-Prot entries
6. UniProt databases
30 January 2015
6
ENA/GenBank/DDBJ, Ensembl, VEGA, RefSeq, other sequence resources
UniSave
- Providing entry
version history
UniSave
- Providing entry
version history
7. Talk outline
• Introduction to UniProt
• UniProtKB annotation and propagation
• Data increase and the need for Automatic Annotation
• Automatic annotation systems in UniProtKB
• UniRule Automatic Annotation System
• Complete Proteomes in UniProtKB
30 January 2015
7
12. 30 January 2015
12
Propagation of annotation in UniProtKB
Annotation Propagated
RecName Yes
AltName Yes
Function Yes
Catalytic activity Yes
Pathway Yes
Subunit Yes
Subcellular location Yes
Disease No
Disruption phenotype No
Polymorphism No
Alternative products No
Generalannotation_______
Featureannotation_______
Annotation Propagated
KW Yes
GO Yes
Regions of interest Yes
Active site Yes
Ligand-binding Yes
Processing Yes
PTMs Yes
Ambiguities No
Conflicts No
Natural variants No
Isoforms No
13. Talk outline
• Introduction to UniProt
• UniProtKB annotation and propagation
• Data increase and the need for Automatic Annotation
• Automatic annotation systems in UniProtKB
• UniRule Automatic Annotation System
• Complete Proteomes in UniProtKB
30 January 2015
13
15. 30 January 2015
15
Benefits of Automatic Annotation
• Added value for TrEMBL in the face of rapid data growth
• many species/proteins without published experimental data
• Support for manual curation
• making manual curation of TrEMBL entries for which there is
published data easier
• Correction of misleading annotation in data received from
sequencing centres
• Highlighting of patterns
• knowledge that can be/needs to be propagated across the
databases
• inconsistent annotation e.g. of a protein family
16. Talk outline
• Introduction to UniProt
• UniProtKB annotation and propagation
• Data increase and the need for Automatic Annotation
• Automatic annotation systems in UniProtKB
• UniRule Automatic Annotation System
• Complete Proteomes in UniProtKB
30 January 2015
16
17. 30 January 2015
17
Automatic Annotation in UniProtKB/TrEMBL
We have implemented Automatic Annotation systems based
on annotation rules
•Rules are linked to specific signatures - InterPro
•Annotation rules have
• annotations
• conditions
•Rules are tested and validated against UniProtKB/Swiss-
Prot
•Rules and annotations are updated each UniProt release
18. Automatic Annotation Systems in UniProtKB
System
Rule
creation
Trigger Annotations Scope
SAAS automatic
taxonomy
InterPro
comments, KW all taxa
UniRule
(Rulebase/HAMAP/
PIRNR/PIRSR)
manual
taxonomy
InterPro*
proteome
property
sequence
length
protein names,
comments,
features, KW,
GO terms
all taxa
30 January 2015
18
* Flexibility to create custom signatures and submitted to InterPro as required
19. Principle of an Annotation Rule Creation
30 January 2015
19
annotated Swiss-Prot entries rule TrEMBL entries
extract common
annotation
propagate
taxonomic nodes
Interpro entries and member signatures
proteome properties
sequence length
TrEMBL entries remain in TrEMBL, but offer more (predicted) annotation
20. 30 January 2015
20
SAAS – Statistically Automatic Annotation System
• Automatically generated annotation rule system to
supplement the labour intensive UniRule system
• Employs a C4.5 decision-tree algorithm to find the most
concise rule
21. Talk outline
• Introduction to UniProt
• UniProtKB annotation and propagation
• Data increase and the need for Automatic Annotation
• Automatic annotation systems in UniProtKB
• UniRule Automatic Annotation System
• Complete Proteomes in UniProtKB
30 January 2015
21
22. 30 January 2015
22
UniRule Automatic Annotation System
• Manually created/curated rules of varying complexity:
annotation varies from simple Keyword attribution to
complete annotation
• Sources for rule creation
• automatically generated SAAS rules as input
• literature based curation of characterised families – as a
potential source for creating new signatures for a specific
functional group
• also …
23. UniRule - conditions used to created a rule
Conditions (can be positive or negative)
•Taxonomy
•InterPro entries and member signatures
•Subcellular location e.g. organelles
•Proteome properties e.g. photosynthetic
•Sequence length
30 January 2015
23
24. UniRule – UniProtKB annotations defined in a rule
Annotations
•Description lines
• Protein names
• EC numbers
•Gene names
•General annotation (comments)
•UniProtKB Keywords
•GO terms
30 January 2015
24
29. Talk outline
• Introduction to UniProt
• UniProtKB annotation and propagation
• Data increase and the need for Automatic Annotation
• Automatic annotation systems in UniProtKB
• UniRule Automatic Annotation System
• Complete Proteomes in UniProtKB
30 January 2015
29
30. How does UniProt define a Complete
proteome?
• A complete proteome consists of the set of proteins
thought to be expressed by an organism whose genome
has been completely sequenced.
30 January 2015
30
31. Status of complete proteomes in UniProt
• Longstanding project, 2902 proteomes that are spread over the
entire taxonomic range
• Archaea
• Bacteria
• Eukaryota
• Viruses
• Capture of “Complete proteome” data is a mixture of automatic and
manual procedures
• Aim is to provide a set of UniProtKB entries that define the proteome
30 January 2015
31
32. Human complete proteome
• First draft of the complete human proteome available in
UniProtKB/Swiss-Prot in September 2008
• The first mammalian proteome to be annotated
• Representing approximately 20,000 putative protein-coding
genes each represented by one canonical sequence
30 January 2015
32
33. Other complete proteomes
Human not the only organism to have its proteome annotated
•Sus scrofa (Pig) – 19,576 entries
•Gallus gallus (Chicken) – 21,622 entries
•Mus musculus (Mouse) – 46,656 entries
•Arabidopsis thaliana (Mouse-ear cress) - 32,521 entries
30 January 2015
33
34. Challenges of proteome data
• How to define a complete genome, what is complete? Does it have a
complete set of gene model annotations?
• Track any changes in the genome annotations and the impact on
UniProt
• Gather all proteomes available, develop import pipelines to improve
species coverage, current sources include:
• INSDC species
• Ensembl species
• UniProtKB also define a subset of the Complete proteomes as being
'Reference proteomes'.
• Complete proteome of a representative, well-studied model organism or
an organism of interest for biomedical research.
30 January 2015
34
37. Closing remarks
• Manual annotation cannot keep pace with current or future
rates of growth of UniProtKB so there is a need for automatic
annotation
• UniProtKB currently uses two automatic annotation systems
referred to as SAAS and UniRule
• Automatic annotation of TrEMBL is refreshed and validated
using UniProtKB/Swiss-Prot as a reference, each UniProtKB
release
30 January 2015
37
38. Closing remarks
• UniRule – manually annotated rules
• annotation varies from simple keywords to full annotation
• starting from SAAS rules, InterPro signatures, literature-based
curation of protein families
• possibility to create custom signatures for InterPro
• Evidence attribution - users to determine the composition
of the rule behind predicted annotation
30 January 2015
38
39. Closing remarks
• Requirements for completed proteomes
• Completely sequenced genome
• Good gene prediction models
• Good quality transcriptome/proteome data
• Proteins are mapped to genome
30/01/15
39
40. Acknowledgements
• UniProt group at the EBI is led by Claire Odonovan and Maria
Jesus Martin, part of the PANDA proteins group led by Rolf
Apweiler
• UniProt group at PIR, Georgetown University is led by Cathy
Wu
• UniProt group at SIB (Geneva/Lausanne) is led by Ioannis
Xenarios and Lydie Bougeleret (heirs to Amos Bairoch, left
2009)
• Thanks also to all curators, developers and support staff at
all three sites
30 January 2015
40
41. Funding
• National Institutes of Health (NIH)
• European Commission (EC)'s SLING
• Swiss Federal Government through the Federal Office of
Education and Science
• GEN2PHEN
• MICROME
• National Science Foundation (NSF)
30 January 2015
41
Notas do Editor
Good morning. My name is Wei Mun Chan and I am a UniProt curator working at the European Bioniformatics Institute, in Hinxton, UK.
In the following session I will talk about Automatic Annotation as it is used in UniProtKB, focusing on UniRAule, and I will subsequently talk about Complete Proteomes in UniProtKB.
Firstly, I will talk about the background to UniProt.
I’ll then go on to describe UniProtKB annotation and propagation of this annotation.
Moving on, I will talk about the growth of data in UniProtKB and the need for Automatic Annotation to cope with this data.
I’ll then describe the Automatic Annotation Systems we use in UniProtKB.
And to end, I will describe UniProtKB Complete Proteomes.
The UniProt Consortium was formed in 2002 – together to provide the Universal Protein Resource UniProt.
This is a screenshot of the UniProt Resource, which is available at the URL www.uniprot.org.
This is the entry point to the various resources offered by UniProt.
The mission of UniProt is to provide the scientific community with a comprehensive, high-quality and freely accessible resource of protein sequence and functional information.
UniProt provides several databases including:
UniSave – which contains the entry version history for UniProtKB entries.
UniRef – sequence cluster database.
UniMES – stores metagenomic and environmental sequences.
UniParc – Sequence archive
The centre piece of UniProt’s activities is the UniProt KnowledgeBase UniProtKB, which contains the Manually reviewed UniProtKB/Swiss-Prot entries and Unreviewed TrEMBL entries.
Unreviewed UniProtKB/TrEMBL entries are manually reviewed by experts and upon integration/become reviewed UniProtKB/Swiss-Prot entries. Automatic Annotation procedures then feed back into the UniProtKB/TrEMBL database.
Now to give you a brief overview of the information annotated in a UniProtKB entry.
Take the human protein Ky-nu-renin-ase as an example. This is a screenshot taken of the top section of the protein entry as it is represented in the UniProtKB database.
As you can see the protein has been manually annotated, Reviewed entry and therefore resides in the UniProtKB/Swiss-Prot database.
Within this entry, we provide a Recommended name, Alternative name, Gene Symbol, Organism and Taxonomy.
In addition, within the entry we also provide further information that can be attributed to the protein in General annotation comments section. Here we collate valuable protein characterisation information.
Annotation in UniProtKB also includes the annotation of Ontologies.
And this includes the annotation of UniProtKB-specific Keywords and also the complimentary Gene Ontology terms.
We also provide sequence feature annotation – where we describe for example how the molecule is processed, regions, sites within the sequence. Also amino acid modifications and natural sequence variations.
This slide captures the graphical representation of sequence features in UniProtKB.
As part of our manual annotation efforts we also propagate annotation from proteins where the annotation has been demonstrated experimentally.
We restrict the propagation between closely related species, or if appropriate also to distantly related species where the proteins are well-conserved.
This table lists some of the categories of annotations that we do or do not propagate between UniProtKB entries.
This graph illustrates the disparity between the growth in the number of:
Manually reviewed UniProtKB/Swiss-Prot entries represented by the red line.
Unreviewed automatically generated UniProtKB/TrEMBL entries represented by the green line.
Thus, our manual curation efforts is being outpaced by the growth in the unreviewed data. There is therefore a need to enrich these unreviewed UniProt/TrEMBL entries we annotate.
How. Through the use of Automatic Annotation.
What are the benefits of Automatic Annotation?
Firstly, to enrich unreviewed entries in TrEMBL in the face of rapid data growth. There are many species/proteins without published experimental data.
Secondly, to support our manual curation efforts. Making manual curation of unreviewed TrEMBL entries for which there is published data easier.
Thirdly, to correct misleading annotation in data received from sequencing centres.
Fourthly, to help highlight patterns. Recognise knowledge that CAN or NEEDs to be propagated across the databases. Also to help spot, inconsistent annotation e.g. of a protein family.
How is Automatic Annotation in UniProtKB/TrEMBL accomplished?
We have implemented Automatic Annotation systems based on Annotation rules.
Rules are linked to specific signatures – InterPro.
Rules are associated with – annotations and conditions.
Rules are tested and validated against UniProtKB/Swiss-Prot.
Rules and annotations are updated each UniProt release.
UniProtKB currently employs two automatic annotation systems – referred to as SAAS and UniRule.
Recognises common annotation belonging to a closely related family within UniProtKB/Swiss-Prot
Transfer of common annotation to related family members in UniProtKB/TrEMBL
The manually curated UniRules annotation system – incorporates the HAMAP, RuleBase, PIRNR/SR systems.
This slide shows a simplified overview of how Annotation rules work in UniProtKB.
Rules are built by a combination of specific conditions (grey box) and associated with common annotation from UniProtKB/Swiss-Prot entries.
The rules identify/extract common annotation present in UniProtKB entries (represented by the highlighted red grids) and then propagate these annotations to unreviewed TrEMBL entries.
Thus, the annotationally enriched TrEMBL entries remain in TrEMBL.
The first of the Automatic Annotation Systems used in UniProtKB is:
Statistically Automatic Annotation System abbreviated SAAS.
Supplements the labour intensive manual curation of rules in the UniRule system.
C4.5 algorithm uses entropy gain to find most concise rule.
Manually created/curated rules of varying complexity. Simple UniProtKB Keyword to complete annotation.
Sources for rule creation – roughly categorised into four categories:
1) Automatically generated SAAS rules as input.
2) Literature-based curation of characterised families – as a potential source for creating new signatures for specific functional group
And also:
3. Conditions.
And 4. Annotations.
This is a screenshot to show how links to UniRule rules is represented in a UniProtKB entry.
We can see that the rule of interest is UniRule/RuleBase RU003777.
Information that has been propagated to this UniProtKB/TrEMBL entry by this rule is followed by the RuleBase evidence tag.
The same evidence tag also appears associated with General Annotation and Ontologies.
Clicking one of these tags/links allows us to view the predicted annotations and the rule itself.
On a left we see the predicted annotations. And on the right we see the conditions used to build the rule (conditions can be positive or negative).
On a left we see the predicted annotations. And on the right we see the conditions used to build the rule (conditions can be positive or negative).
So what are some of the challenges UniProt faces when dealing with Complete Proteome data?
How do we define a complete genome, and decide what is complete?
For example - does it have a complete set of gene model annotations?
There is a need to track changes in genome annotations and how these changes impact on UniProt.
Gather all proteomes available,