Reference Genome Meeting Minutes April 2008
- 1 April 20, 2008
- 1.1 Annotation Progress
- 1.2 Annotation Pipeline, Part 1: Generation of protein sets (Suzi Lewis)
- 1.3 Software Update
- 1.4 Annotation Pipeline, part 2 (Suzi Lewis and Judy Blake)
- 1.5 Consistency within experimental annotations (Pascale)
- 1.6 Using Textpressor (Kimberly Van Auken)
- 1.7 Annotation of isoforms (Harold Drabkin)
- 1.8 Anything else to add?
- 2 April 21, 2008
- 3 Orthology, Paralogy, and GO Annotation
- 4 Review old action items
April 20, 2008
Annotation Progress (Mike Cherry)
- Number of annotated genes per organism by evidence type (overall): If Compare graphs for Sept 2007 and Apr 2008 see that over all size and size the same, but IEA decreasing.
- What is effort/person?
- X-axis is absolute number of genes, which doesn't reflect differences in genome size.
- Number of annotated genes per organism by evidence code for Reference Genome project: the majority of genes have experimental evidence codes
- Graph needs outline that indicates "no ortholog". This allows a comparison of the genes present or absent in the reference genome genomes. It will also show which organisms are lagging behind.
- Number of annotations as a metric? would give a different view of the progress, but too variable b/c of differences in depth of knowledge in different organisms, different areas of the ontology.
- View progress between Sept 2007 and April 2008 as a % change. Can see that everyone has doubled experimental annotations, although it doesn't show the starting number of annotations.
- Need to discuss which metrics we want to track and why. Need consistent measures across groups.
- How annotations change over time lets you see whether groups are still engaged in the process.
- Would be useful to have a display that shows how much is known about these genes. Some of this information will come from Chris's reports.
Annotation Progress (Chris Mungall)
- distance to leaf (shows average number for all genes)
- didn't change between Jan 2006 to Sept 2007
- consider breaking down by the 3 ontologies, also show % of length to leaf
- information content
- a quality control measure
- coverage (# of nodes covered per gene)
- as there is more information about a gene it will have more coverage
- can there be too much coverage?
- publications per gene
- GO terms per gene
- For each of these metrics: what is appropriate range for each category? need a sense of the scale, perhaps express X-axis as a % rather than an absolute number.
- distance to leaf (shows average number for all genes)
- Reference Genome Reports
Annotation Progress: Discussion of other ideas for measuring progress
- A measure that shows progress made in curating the experimental literature for reference genes in reference genomes. This is an aim of the grant. Can determine number of publications annotated.
- A measure of time spent (% effort) actually doing experimental annotations. Disagreements: Can't do curation w/o ontology development and visa versa. Worried about trying to parse out too much. How to you separate annotation from time spent considering how you do annotations or assessing quality of annotations.
- A measure of the number of genes that have been comprehensively annotated.
Annotation Pipeline, Part 1: Generation of protein sets (Suzi Lewis)
- The issue is determining a procedure to define a coherent set of orthologous proteins. For experimental annotations, want to annotate to isoforms. But for tree building want longest protein produced from a gene. So for ortho sets want a unique protein/gene ID for the "canonical" gene/protein.
- Proposal from Chris Mungall for the Gene Association File: In column 2 will have the ID for the "canonical" gene. Add an additional column (column 17) to hold the ID for the thing that was annotated (protein/gene/transcript). Column 17 must match column 12 (SO type).
- Background for the proposal
- Currently there is heterogeneity in column 2 of Gene Association files (see Annotation of alternate spliceforms, some groups have protein IDs, some groups have gene IDS, some groups have a mixture of both.
- Add a column that is always for a gene. A gene is a "concept", it's a lumping term that reflects biological reality. It provides the link we want.
- Alternate proposal from Rex Chisholm: Keep column 2 as it is now, the ID for the thing that was annotated. (In a perfect world, this would be the gene product.) Keep column 12 it is is now, referring to column 2. Add column 17 for the ID for the "canonical" gene that codes for the product that was annotated.
- Have to look at how any change will affect our users. What do users expect to be in column 2? they expect canonical ID, but it isn't always the case.
- Decision: Most groups in favor of the proposal of making column 2 the canonical ID and the column 17 the ID for the thing that was annotated.
- What should column 12 refer to?
- Decision: Column 12 should point to column 17, which means that column 17 must be filled in; it can't be left blank and inferred from column 2.
- Notifying users. Before change is implemented, should it be discussed with a few users? Need a pushout list to identify users of changes/updates.
- The header of gene association file should state this file contains annotations for x out of total number of genes estimated in this organism.
- Background for the proposal
- Proposal for the gp2protein file: For every canonical gene ID in the GAF there will be an associated canonical protein ID in the gp2protein file.
- Background. Still need gene to protein association. This should be in a separate file from the gene association file. The gp2protein file seems the logical place to have this association.
- What about those cases where gene has been annotated, but there is no known protein sequence associated with it. Should there be a blank in the gp2protein file? or should the gp2protein file have information that the gene is "uncloned" or codes a functional RNA? Don't want to overload the file (putting non ID information in an ID column). If needed, should make a separate file or find other ways of dealing with the blanks. Can generate report that gives type from column 12.
- If gp2protein file has only canonical protein IDs, how do you get information about other protein IDs (column 17)?
- syntax for gp2protein file: Enter protein accessions as UniProtKB:xxx or NCBI:xxx
- Gene Association File (GAF)
- column 2 is canonical gene ID
- column 17 is thing you are annotating (always required)
- column 12 matches column 17 and contains SO ID's
- gp2protein file:
- includes complete gene index (except for pseudogenes and transposons)
- column 1 is canonical gene ID
- column 2 is accession for sequence of longest form of protein from UniProtKB: or NCBI:
- Action Items
- update documentation
- write notice of changes to users
- individual data providers make sure that their input matches
- software changes as necessary
- add header to gene association file
- syntex of gp2protein file will be provided by Mike and Chris
Demo of RefGenome tracker interface (Siddhartha Basu)
- RefGenome tracker interface (database to replace current google spreadsheet)
- For programmers:
- add box for taxon id to the "add target" entry box
- add column with MOD id so curators can link to the MOD rather than NCBI
- Time frame?
RefGenome Graphs (Mary Dolan)
- Comparison matrix of GO terms across organism
- Suggestion: For ISS include the "with" information
AmiGO (Seth Carbon)
- Summary table of genes in the RefGenomes List
- Want feedback on summary table and visual graphical displays. Are there other types of visual displays that people need?
- AmiGO will gradually be moving to this structure.
- Cross Products
- Example TAZ gene, various annotations related to heart in different organisms, but couldn't see connections in the graph. Working on version of graph display to show these connections.
Community Annotation at GONUTS (Jim Hu)
- Demo of "Create New Gene Page"
- Have a webservice that can be used to connect AmiGO and GONUTS by identify pages in GONUTS that have been annotated by a human being
- two possible types of input: small, individual annotations; bulk sets of predictions
- could you use it for getting input on IEA annotations?
- will increase input by making it easier for people to provide input
- try connecting Cardio and Immunology pages to GONUTS
Annotation Pipeline, part 2 (Suzi Lewis and Judy Blake)
- Is there consensus about the steps shown in File:Ref Genome annotation pipeline2008Mar31.pdf?
- Need to discuss how the "focal-sets" will be determined
- Step V: Changed to "Curators add/remove proteins to/from the "focal set" based on dialog and agreement"
- What is the purpose of the focal set? defined at meeting in Princeton, to be able to say that these products have been experimentally annotated across the reference genomes
- Rex: Agrees with procedure but wants to emphasize that it can't be written in stone. There has to be option for future discussion changes in the procedure. Suzi: Yes, in light of further knowledge the procedure may change and people involved have to be open-minded and leave behind their preconceived notions.
- Rex: Want to capture both depth and breadth (annotations to as many genes as possible based on exptl. annotations and also ISS).
- Judy: Inferential annotations. how do you transform experimental annotations in one organism to inferential annotations in another organism? what measures are useful? what about large family sets?
- Suzi: QC is something that happens during the entire process, not just at the end of the process. Will be useful to think about QC at each of these steps.
Consistency within experimental annotations (Pascale)
* Tricky terms Misused_terms * Outliers * Val (co-occurences of annotations)
- Annotation errors
- transient localization vs long-term cellular localization
- secreted protein annotated to secretory process
- IMP evidence code used for results from high through-put experiments
David: Once a focal set is annotated, send one of Mary's graph to someone who has published a lot of papers on that gene/protein and ask them if there is anything wrong.
Fixing errors: should there be a systematic effort to review earlier annotations?
Develop SOPs to prevent future errors.
Are there automatic checks that can be done to identify anomalies to be reivewed?
Pop-ups to warn curator?
At end of day, look at totality of annotations in other organisms and then review the annotation of your gene.
[ACTION ITEM] QuickGO cocurrency terms, maybe this should be in AmiGO?
How do we continue to study the process in order to improve it?
Judy: Have a list of misused terms and comments made about them. Where can this information be put to make it more visible?
Page needs to provide more information about how the term was misused.
[moving towards an ACTION ITEM] Is review by a human the only way to check the usage of frequently misused terms?
growth, cell growth, cell cycle, cell proliferation frequently used interchangeably
- Part terms.
- Need to use them, would be helpful to have SOPs.
- Possible way to identify outliers (Val)
SLIM by SLIM matrix to review intersections of different cellular processes and look for unexpected intersections which may identify possible errors
try applying to function and component terms
outline cells that you expect to be empty
[action item] generated automatically from AmiGO database rather than each refgenome group doing it themselves
spot checks have to be built into the process, need to build different ways of looking at quality control
suggestion: collect tricky terms, run reports, and email groups asking them to review their annotations. If annotations are correct, can drop term from the tricky list.
problem in ontology systematic problem xx?
can you find systematic errors by looking at co-occurrence of misused terms with particular paper?
David: often thought that if look at set of annotations from a particular paper can get more information about the gene
Amazon shopping cart model: 90% of annotators who used this term
people should be on lookout for misused terms and add to list along with explanation
need way of notifying people that something was added
also need regular process for resolving these
a) software for generating comatrix b) buddy annotation c) categories of problems, not just endless list of problems d) regular assessment
adding comments in OBO-edit
- additional areas where the groups aren't consistent and need to be discussed
- discussions at annotation camps are very helpful
annotation consistency test: summary was that there was no consistency does this mean that refgenome project is doomed? no, set included people who had never annotated before and only small number of people that worked on the organism
Using Textpressor (Kimberly Van Auken)
Textpresso for GO annotation key features:
- search through fulltext
- in addition to keyword searches, have category searches based on groups of related words
Wormbase uses Textpresso
- get PDFs
- convert to text
- marked up by Textpresso
Textpresso for wormbase curation presentation. Example from looking for P granule annotation. Most papers were P granule mislocalization in mutant. Need relevance markup. Hired a student to go through papers and mark up localization based on antibody staining. 219 pubs, 1400 sentences. Used curation form that divides sentences. Student checked yes or no for relevance. Compute word frequency histograms. Single words or phrases. Single words worked well and are more efficient than phrases. Created Textpresso categories
- Cellular component: Adherens junction, nuclei,
- Verbs: localizes to, accumulates
- Other: ... missed this.
Metrics: precision and recall. First generation categories 75% precision 40% recall. Could get 80% of the known annotations, thanks to info redundancy. Building second generation categories. Curation pipeline.
- Keyword - Ce protein name
- Look for match in 3 categories
Returns matching sentences in documents. Can browse sentences in context. Sentences get a score. Interface - 3 columns: Protein, Textpresso match terms, GO terms from relationship index.
- working on problem of how to identify new associations from those that have been reported before, e.g. commonly used markers
- how much does this increase the efficiency of curation? can't answer right now because still testing
- how does it affect your annotation? how do you know what you're missing?
- current pipeline is just for cellular component terms, but think it will be amenable to function terms, haven't thought about application to process terms
- editor tool that will let you customize category terms
Annotation of isoforms (Harold Drabkin)
Problem: representation of multiple proteins forms of a gene generated by: natural variation, alternate splicing, etc. This is a problem because the isoforms may have different functions, localization, processes.
MGI database has place for notes that can be mined. gene product field: annotation was for a specific isoform
different types of information that can be captured: a) anatomy, specific cell type b) specific term in evidence code ontology, included transcript & protein ID c) specific evidence code, cell type, product ID
File listing individual isoform annotation contains information that wouldn't show up in GO annotation b/c MGI has
Challenges What genes have isoforms? do the isoforms have ids? do the isoforms actually exist Other: function protein domains/fragments; modifications; both of these related to a single isoform by derivation
Strategies for back populating 3700 uniprot records with isoforms, look for those that have references that were used for GO have 1693 markers with more than one NM_ record papers with isoform or alternating splicing in title or abstract
Example: Notch1 full protein, transmembrane binds extracellular ligand intracellular domain is cleaved to give NICD, which goes to nucleus, binds RBJ-1 and functions as txn co-activator
Focus: transporters SLCxyz, ABCxyz interleukins Others?
How to represent xx? current practice does not allow dual annotation column 17 may solve this problem
Anything else to add?
keep protein isoforms and multiple transcript problems separate
April 21, 2008
Judy reviewed some aspects of mutliple orthology sets - a lot of different resources, and linking between them.
- Mammalian orthology sets at MGI
- Homologene -nice but doesn't include all reference genomes
- TreeFam - maximum likelihood-based
- PIR homeomorphic protein superfamilies
why do we need another resource (i.e., Kara's)? the above tools are not comprehensive, and don't start with gene index. [discussion regarding these reasons: why can't the above tools incorporate our protein sets? who do we need to contact/work with?]
- we spend a lot of time determining what genes are/not in an ortholog set - is that the best use of our time? should we use existing resources?
Judy showed some slides from Kara.
- Kara had offered to address some of these concerns by doing a PPOD run specifically for the refGenome project: PPOD refGenome database
Orthology, Paralogy, and GO Annotation
- goal of refGenome project is to identify genes in reference genomes that have same or similar functions, so can do comprehensive curation simultaneously
- ortholog = same gene in different organisms separated only by speciation
- orthologs can have different functions
- paralogs can have same functions
What is an ortholog cluster?
- Algorithms make slices through protein trees based on some combo of evolutionary rates and history of duplications/speciation
- They make arbitrary calls based on calculations; different algorithms will do this differently.
- One must still investigate the experimental biology, and make educated judgments
- It can still be quite useful to have comprehensive curation of related genes, even if they don't technically fall into the ortholog cluster (can be "fruitfully" annotated at same time
Tree visualization tool for Ref Genomes - in development
- Pre-computed searchable library of gene trees - modifiable based on curator feedback, includes outgroups
- Visualization tools - trees labeled with GO annotations
- Homology annotations supported by tree evidence, available to scientific community
- HMMs to allow other genome projects to infer GO annotations
Now that we have all protein IDs together, Paul can have a run completed by July (parsimony-based trees)
- ACTION ITEM: Paul will grab gp2protein files on May 1st and begin his run
We will prioritize genes that are present in Kara's data in all 12 reference genomes - there are 153 of these orthogroups
- we will do 20 genes/month starting at top of alphabetical list (by human gene name)
- we will place more focus on experimental literature and less focus on inferential annotations
We will continue to do QC as one/month/curator (at most, if it takes longer that's ok, must balance with other work tasks)
- using refGenome sourceforge tracker
Review old action items
[ACTION ITEM]: Done. (everybody): We will add categories of genes to annotate in addition to ‘disease genes’. We will choose five genes from each of the following four groups: 1. diseases 2. biochemical/ signaling pathways 3. bleeding edge list: 4. conserved genes/unannotated genes. This will be done on a rotation basis from all databases.
[ACTION ITEM] (Val): Provide the list of 207 genes conserved between pombe and human with no annotation/information
[ACTION ITEM] (Jim): Provide the set of conserved genes found by InParanoid that are conserved in all 12 species (660 or so); we might want to prioritize this list by ascending order of number of annotations to target unannotated genes (who can do that?) DONE, see 'Suggestions' spreadsheet (look for the "conserved Hs-Ec" sheet)
[ACTION ITEM] (Ruth): send the HGNC list of genes with few annotations
[ACTION ITEM] Done Amelia (web page), Susan, Rex, Petra (content), work on web presence
[ACTION ITEM] In progress : contact/meet with people who have made tools for orthology determination to see if they can help us (that possibly includes re-running the analyses using the most recent set of sequences and proper IDs; Compara: Emily? Homologene: Judy? TreeFam in paranoid others?
[ACTION ITEM]: In progress, Kara: run the P-POD over the full ref genomes set? analysis on the ref genome data set. Need computational pipeline with existing resources. Currently takes 3 weeks to do 8 species all v all. Goal was set for February 2008 to include all ref genome sets.
[ACTION ITEM] In progress Re-calculate with is_a only paths (Chris)
[ACTION ITEM] In progress, Re-calculate with experimental codes only; generate several versions of the data classified by different evidence codes?
[ACTION ITEM] In progress (Chris) Provide such reports on a regular basis
[ACTION ITEM] Done (Chris): pull out ND annotations and report to each group, see Reference_Genome_Database_Reports
[ACTION ITEM] (Software group) In progress: Continue working on development the curation tool
[ACTION ITEM] (Mary Dolan, Software group) In progress Continue working on the best way to display annotations graphically.
[ACTION ITEM] (Pascale Gaudet) - paper -Topics: talk about goals, process, curation priority, ortholog finding, interesting biology (outliers reflect mistakes in annotations or interesting differences in the biology), benefits (improve annotation consistency and ontology quality)
[ACTION ITEM] (Mike Cherry, who else): Organize a reference genome annotation camp, possibly in spring or summer of 2008.
[ACTION ITEM] (Judy Blake) Contact NCBI/NLM/OMIM to link to reference genome genes
[ACTION ITEM](Pascale Gaudet): Document in SOPs:
[ACTION ITEM] Rejected, (Pascale Gaudet): Add to RefGenome curation practices SOPs: please enter your unique gene_id in the google spreadsheet (Makes it easier to parse)
[ACTION ITEM]: Rejected (Pascale Gaudet) Generally, provide guidelines for filling the google spreadsheet (IDs, where to put notes, etc)
Counting papers, assessing completeness/comprehensive annotation status Ruth Lovering Ideally, the completeness of annotation of a gene happens when all papers referring to that gene have been read by a curator (whether or not those papers provided GO new annotations). In practice, however, curating all literature for a gene in only feasible when there are relatively few papers (maximum 5-10). Thus, although we were compiling the number of papers published for a gene, the number of papers read, and the number of papers used for annotation (columns LMNO of the Google spreadsheets), we eventually realized that this number was not always representative of the depth of curation (for example, 20 papers curated out of 1000 is only 5% of the literature, but the gene might be curated to great detail; while 1 paper out of 2 for another gene gives 50% and is probably under-curated). We all agreed that the measures Chris presented are both more meaningful and easier to get (since they can be generated automatically), and therefore we will stop keeping track of the papers.
[ACTION ITEM](Pascale Gaudet): Document in SOPs Another factor we have been tracking is when a curator judges that the curation of a gene is ‘comprehensive’, that is, that is accurately represents the biology (irrespective of the number of papers available or read). This appears in the spreadsheets. The guideline is that when there are few papers, all papers should be read; when there are many (a curator can judge what is too many), then a review should be read to find the important primary literature and decide what information needs to be captured. We don’t keep track of whether or not reviews have been read. Wormbase uses textpresso (PMID 15383839), that helps ensuring curators do not overlook information. The ‘comprehensive’ curation status doesn’t get invalidated when a newer paper is published; however, curators may (and are encouraged to) update the date when the newer literature is curated.