PomBase conventions for improving annotation depth, breadth, consistency and accuracy
Download as PPTX, PDF1 like265 views
Pombase employs a variety of annotation conventions and quality control measures to enhance the depth, breadth, consistency, and accuracy of its data annotations. Critical to this is the use of precise definitions and specificity in annotations, as well as addressing errors and inconsistencies through community curation and automated processes. The goal is to ensure robust, clear, and scientifically accurate annotations that facilitate better data mining and analysis.
PomBase conventions for improving annotation depth, breadth, consistency and accuracy
- 1. PomBase conventions for improving annotation depth, breadth, consistency and accuracy
- 2. Annotation numbers are important …but numbers aren’t everything….. • Use of annotation for data-mining and data-analysis is limited by errors, inconsistencies and omissions. • PomBase uses a combination of annotation conventions, to improve information content (annotation coverage, specificity and redundancy), and QC mechanisms to identify possible annotation inconsistencies and errors. • In combination these mechanisms address many recurring annotation issues.
- 3. 1. The definition is critical All ontology terms have a “fixed” definition • If a definition is misleading or incorrect its meaning cannot be changed. To fix the term is obsoleted and annotations are migrated. • This makes annotations very robust to ontology changes. If a term needs to be repositioned the annotations remain correct . • We annotate to the definition, not the term name. Always check the definition.
- 4. 2. Improving annotation specificity • i) Consider descendant terms • ii) Veto use of uninformative terms
- 5. 2i. Consider descendants Annotate as specifically as experiment allows and be unambiguous about the biology • regulation: positive or negative? • translation: cytoplasmic or mitochondrial? • transport: of what? to where? how? • chromosome segregation: mitotic or meiotic? If the available terms are insufficient, request a more specific term
- 6. • For a carboxylic acid carrier “carboxylic acid transport” looks initially OK • However “transmembrane transport” is not explicit here… Carboxylic acid might be transported in other ways… 2i. Consider descendants e.g.
- 7. More specific annotation can provide additional detail e.g. • substrate, • type (transmembrane), • sometimes directionality Additional parents increase the information content as annotating indirectly to more terms. 2. Consider descendants e.g.
- 8. 2. Veto use of non-specific terms Identify the set of ontology terms where more specific annotation should be possible (more biological detail) Examples: • e.g. cellular process (which one?) • e.g. translation (cytoplasmic? mitochondrial?) • e.g. transport ( of what? to where? ) Some GO terms are already flagged as not for manual annotation. Review and improve annotations to vetoed terms PomBase blocks 1298 upper level GO terms for direct annotation (<200 violations)
- 9. 3. i) Missing parents Original arrangement 3. Improve the ontologies
- 10. 3i. Missing parents These process annotations were originally in different branches of the ontology, so all annotations were required
- 11. New arrangement: 3i. Missing parents
- 12. 3.i Missing parents Collapsed 6 processes to 2. Exactly the same information content Less redundancy, easier for users to interpret annotation
- 13. 3.ii Report incorrect parents AKA “True Path Violations” or “TPVs” For example protein maturation --protein processing (part_of) ----proteolysis (part_of) (not all proteolysis is processing or maturation)
- 14. 4. The power of Annotation Extensions Provide additional specificity for a GO annotation e.g. • Target gene (kinase substrate, TF regulation target) • Location of a function • Localization dependencies (protein A localizes protein B) • Spatial and temporal aspects of processes, functions, locations (cell cycle stage of occurrence) • ADD an example of a gene product specific AE See: Huntley et. al. A method for increasing expressivity of Gene Ontology annotations using a compositional approach. PMID:24885854
- 15. cyclin-dependent protein serine/threonine kinase • has substrate fkh2 involved in negative regulation of conjugation with cellular fusion • directly inhibits srw1 involved in positive regulation regulation of G1/S transition • has substrate drc1 involved in positive regulation of mitotic cell cycle DNA replication • has substrate cdc18, orc2 involved in negative regulation of DNA replication during mitotic G2 phase • has substrate xlf1 involved in negative regulation of double-strand break repair via nonhomologous end joining, during mitotic G2 phase • has substrate rap1 involved in negative regulation of mitotic telomere tethering at nuclear periphery during mitotic M phase • has substrate hcn1 during mitotic M phase • has substrate cut3 involved in positive regulation of mitotic chromosome condensation during mitotic metaphase • has substrate mde4 involved in correction of merotelic attachment, mitotic during mitotic metaphase • has substrate, nsk1, involved in negative regulation of attachment of mitotic spindle microtubules during mitotic metaphase • has substrate mde4,cut7 involved in negative regulation of mitotic spindle elongation during mitotic metaphase • has substrate klp9 involved in negative regulation of mitotic spindle elongation during mitotic anaphase A • directly inhibits clp1 involved in negative regulation of exit from mitosis • has substrate byr4 involved in positive regulation of septation initiation signaling • directly inhibits dis2, • has substrate rum1, crb2, sds23 Link function (cyclin-dependent-kinase) to target genes, processes, and temporal information 4. Annotation Extension e.g. cdc2
- 16. Alternative (human CDK1): Not scalable or maintainable
- 17. 4. Using AE for effectors • Reciprocal of the extension (automated) called “target of” • Collects known “upstream effectors” on cdc2 page
- 18. • We can use effector substrate connections to generate networks (interaction, metabolic, regulatory) • Provide directional links to support pathway reconstruction 4. Using Annotation Extensions to generate networks/pathways sty1 cmk2 srk1 rum1 atf1 srk1 gsa1 gpx1 ntp1 sro1 ish1
- 19. 4. Automated AE networks e.g. 44/59 connected in automated network based on annotated connections within “regulation of G2/M transition” (fission yeast) (Network for each GO slim category from the slim page)
- 20. 5. Suppress redundant IEA annotation • PomBase pipelines filter redundant IEA (Inferred from Electronic Annotation) evidence • Removes >90% of IEA (because an existing manual annotation exists)
- 21. 5. Suppress redundant IEA annotation 13 annotations are reduced to 4 Same information, fewer terms
- 22. Incorrect annotations are more easily spotted Mis16 is not involved in ‘chromatin modification,- > fix mapping 5. Suppress redundant IEA, QC of mappings
- 23. Missing parents in ontology more obvious “inorganic anion exchanger” should be an ‘ancestor’ of GO:0005452, to suppress the IEA as redundant 5. Suppress redundant IEA, QC of ontology (SPBC543.05c)
- 24. 5. Suppress redundant IEA annotation • >40,000 fission yeast IEAs available. • PomBase filter 36000 redundant, retain 4000 (IEAs are at least 90% accurate if manual correct). • It is easier to evaluate the remaining IEA’s to identify/fix anomalies Reducing IEAs over time
- 25. 5. Suppress redundant IEA • More concise view with zero loss of information • IEA mappings derived from a single experiment/publication can be interpreted as proof by repetition and make weak EXP data appear multiply supported/acceptable • Fewer annotations, easier QC of remaining IEA’s Q “Why isn’t an IEA covered by manual annotation?” Either: 1. Incorrect mapping 2. Missing parent in ontology 3. Missing annotation -> find supporting evidence and annotate manually (EXP or ISO) (PomBase also filter NAS/TAS/IC)
- 26. 6. Annotate by process (pathway) • Annotating by process rather than “ad hoc” improves consistency and allows ‘annotation gaps’ to be targeted • Process papers more quickly (become more familiar with the field, experimental methods) Become familiar with an area of biology and the techniques used. Don’t need to read the background every time. Recognise phenotypes.
- 27. From PMID:22898774 Regulation of the metaphase/anaphase transition by the MCC, the APC and upstream Signalling Identify obvious missing annotation, for example between complex members 6. Annotate by process or pathway
- 28. 6. Annotate by process or pathway cdc20 proteasome APC separase Cohesin subunit securin Post transition SAC/MCC Can perform QC on processed or components e.g. Use STRING to evaluate outliers (potential annotation errors) Input list “regulation of mitotic metaphase/anaphase transition” Can also ask “are any Complex members missing”
- 29. • We are annotating whole organisms…use a holistic whole annotation approach • Evaluate annotation breadth (coverage) using slims • Evaluate intersections between slim processes 7. Assess annotation at the organismal level
- 30. 7. Evaluate organismal annotation coverage using “slims” • EXP supported BP • ISO/IEA inferred BP ‘unknowns’ • Species specific, no inference possible • Conserved, but unannotated in any species
- 32. 7. Sensible assignments? DNA recombination Periodic check that slim class contents Look sensible
- 33. 7. Monitor unslimmed gene products Note: Exclude biologically uninformative terms like “phosphorylation” or “response to chemical” as these could apply to any real biological role.
- 34. Unknown 830 TOTAL 5054 cytoskeleton org 206 nuclear DNA replica on, recombina on, repair 305 mito c chromosome segrega on 184 regula on of mito c cell cycle 232 10 CELL DIVISION 751 27 cytokinesis 110 0 39 1 46 3 4. MITOCHONDRIAL ORG/EXP 280 4 cell wall org 1303 4 1 MEMBRANES, TRAFFICKING, CELL SURFACE 787 14 lipid met 222 vesicle Mediated transport 324 6 glycosyla on polysacc met 140membrane org 199 75 0 6 74 10 33 0 detox SMALL MOLECULE TM TRANSPORT 288 13 9 0 AA & sulfur met 220 vitamin cofactor met 9 5 nucleo-base/ side/ de met 219 small sugar met 77 CENTRAL MET, ENERGY AND BUILDING BLOCKS 549 Nitrogen 15 25 174 54 3430 other energy genera on 25 23 signalling 404 sexual reproduc ve process 262 (Many intersec ons) Other 290 No intersec ons. Includes adhesion, many proteases, peroxions EXPRESSION 1294 ```` EXPRESSION submod 863 4 1 3 ribosome biogenesis 317 RNA metabolism 772cytoplasmic transla on 249 189 c nucleocyto transport 110 5 34 26 2 Transcrip on 479 32 18 PROTEIN ASSEMBLY/STABILITY 765 protein catabolism & autophagy 251 ubiqui na on 192 63 folding 102 complex Assembly 325 1 3 4 1 7. Visual slim, all pombe proteins
- 35. 7. Evaluate intersections between slim categories Evaluate intersections between processes Many GO processes are rarely co-annotated because they are functionally spatially or temporally distant. For example, would not expect “ribosome biogenesis” to intersect with “vitamin metabolism” We can use this observation to identify potential conflicts using the GO term matrix
- 36. Fission yeast intersections Jan 2012
- 37. Fission yeast intersections March 2017
- 38. 7. Identifies ontology errors (e.g) DNA metabolism and chromosome segregation do not usually intersect Regulation of chromosome condensation should not be a DNA metabolic process
- 39. 7. Ontology error (e.g.) Genes annotated to folic acid metabolism were also incorrectly annotated to amino acid metabolism. Folic acid was classified as an amino acid by CHEBI -> fix, CHEBI, which fixes GO
- 40. 7. Finds incorrect mappings (e.g) Intersect between tRNA metabolism and transcription. Elongator is no longer thought to have a direct role in transcription, mapping removed
- 41. 8. Consider Author intent Think about the biology the author intended e.g. rubidium ion transmembrane transporter/ transport Rubidium ion is used as an assay for K+ transport not rubidium (non-physiological substrate) e.g. Apoptosis (RPS19) Rps19 mutant displayed condensed DNA, a fragmented nucleus and caspase activation - indicative of apoptosis. Since RPS19 has an essential role in ribosome biogenesis apoptosis is likely to be an indirect effect of the disruption of an upstream process translation (i.e. an experimental readout)
- 42. 9. Communication with the author and community curation • Most authors are happy to discuss their publications. If unsure about an annotation ask them. PomBase routinely use the authors as a QC step to refine annotation.
- 43. 9. Community Curation • Most authors are happy to curate their own papers • Co-curation by author and curator improves annotation quality (especially PhD/post doc/recent papers). • 9619 annotations (FTPO/GO/MOD) created by Community from 510 publications (excludes HTP spreadsheet submissions)
- 44. Some example sessions • http://tinyurl.com/q2bgyqv • http://tinyurl.com/p7d979b • http://tinyurl.com/o72bzul
- 45. Very specific annotation is possible because Canto guides the user step by step to construct genotypes and ontology based annotations. “Drill down” to more specific terms is assisted. Prompts are provided for AE of specified types for certain terms.
- 46. 10. Prioritise error fixing • Fixing known errors takes precedence over new annotation.... like critical bugs in code • Even small errors often uncover larger issues, or can fix many problems simultaneously across multiple species. • Prevents propagation of annotation errors
- 47. 11. GO process vs. phenotype • GO annotation should reflect a gene's direct involvement in, or role in regulating, processes or functions. • Phenotypes may indicate that a mutation *affects* a process, but may reflect downstream or indirect effects. e.g. ER membrane defect -> nuclear envelope defect -> chromosome decondensation defect-> defects in next round of DNA replication. • A “DNA replication phenotype” alone is not enough to make a “DNA replication” GO annotation. • Single phenotype is often NOT SPECIFIC FOR A PROCESS.
- 48. Phenotype annotation rules • To make GO annotations based on phenotypes • Ask the question “Is this phenotype or collection of phenotypes specific to this process (usually need detailed phenotypes) Additional data can support GO inference from phenotype (location, orthology), and author intent. (Intersections between processes useful for identifying annotation errors caused by indirect annotation)
- 49. Summary
Editor's Notes
- #2: describe some pombase curation procedures, might be useful to other daabases/curators
- #3: Coverage, genes annotated OR number of different processes for a gene
- #8: Another improtant poitn is that annotations are explicity coupled by using a term which covers both (although this can also be done with extensions)
- #16: Arrange temporally?
- #25: NOTE: we don’t filter redundant EXP annotations, but we do manage this in the display so the term is presented and the source (often multiple) is avaiable in a full view Later we hope to hide higher level EXP annotations
- #29: Complexes cluster together, some genes incorrectly annotated, can work out how they are connected, check appropriate sub porcesses annotated fr complexes, complex annotations are internally consistent etc
- #38: Add error type examples
- #40: CHEbi u= sed to define chemicals in GO
- #42: This isn’t speculative, its the curator using what is known but not explicitly stated, it’s a valid interpretation of the experiment based on what is presented- we are modelling the biology