Entity classes and problems
- 1 Entities marked up
- 2 Classes to add
- 3 Suppressed classes
- 4 Linking issues
- 5 Dealing with ambiguities
- 6 See also
Entities marked up
All WB entities can be viewed here
Entity lists are available here
01downloadModEntities.pl downloads entities
From acedb on spica
From postgres on tazendra
02formSortedLexicon.pl forms the lexicon from the known_entities.
Based on WormBase protocol here, the URL constructor used in the linking script is
Currently we are forming URLs based on public names versus object class ID's. These links are redirected to the correct page, which is the URL based on object ID. By using the public name for the linking instead of the ID, we can avoid problems that may result in links dying due to objects getting merged or killed. In some cases, such as Phenotype, the URL is generated by a mapped WBPhenotypeID.
Object -> source ; web page link
<example object page>
Gene -> postgres on tazendra ; Gene Summary page URL uses public gene name, which is redirected to WBGeneID
Protein -> postgres on tazendra ; Gene Summary page
Allele/SNP -> acedb on spica ; Variation Report page
Transgene -> postgres on tazendra ; Transgene Summary page
Rearrangement -> acedb on spica
Strain -> acedb on spica; Strain Report page
Clone -> acedb on spica; Sequence Summary page
Classes to add
- Human Diseases
- Non-elegans genes - in general as long as the gene name is unique from elegans gene names, and exists in the database, this should not be a problem. One error we ran into is that some names are similar and distinguishing between the elegans gene and the non elegans gene required a correctly formatted species prefix e.g. "Cbr-". Formatting errors are not easily dealt with by automated scripts.
- Anatomy/cell -> acedb on spica with corrections by Raymond ; Summary of anatomy ontology term
This is a problematic class since
- Many cells are named as single letters, ex. "E", "F", "B", etc., which results in many erroneous links
- There are many ambiguities: some cells are also names for common reagents or paper features, e.g., EMS is a cell but also a common mutagen. These names are put on an exclusion list, and if the cell exists in the paper, they are manually linked to the correct page.
- Sex specific anatomy terms such as "Z4.appaap nucleus hermaphrodite" and "Z4.appaap nucleus male" will probably not be an entity that is linked as in general that term is not used in normal writing. The best that we can do is to link such terms to the non-sex-specific anatomy page "Z4.appaap nucleus" where the user can choose the sex-specific term.
- Author stopped linking 04/10/12
This is a problematic class since
- publication names are not consistent
- a single person could have the same name or initials as someone else
- a single person could have multiple aliases.
The linking was limited to the authors on the paper only and verified with the WB person curator to make sure the links were made properly.
Links are formed using the URL pattern:
where $url_encoded_name is the full name of an author with the following modifications:
i) periods after middle names removed and
ii) spaces converted to %20, which is HTML equivalent of a space.
Name: Eve W. L. Chow
Name: I. Russel Lee
(There may be also be other special characters with accents, umlauts, etc., The best thing to do would be to check WormBase and put the correct link, if the author name already exists. Else, the English equivalent of the special character could be visually determined and a new author link can be formed and notified to Cecilia.)
Loss of biological relevance
- Genes within transgene expressions linked to gene pages rather than to transgene pages - gene elements in the transgene genotype is marked up and is linked to the gene page, for example, in the expression eor-1::GFP, only eor-1 is marked up. This is fundamentally wrong as the entire expression should be linked to the transgene page. There are a couple of issues to deal with for this class.
- Not all transgenes are captured by our database. Specifically, Ex arrays and plasmids/cosmids used for injections are not captured since expression patterns for these elements are fleeting, unlike integrated arrays, so are not stable objects to curate.
- In cases where people name the elements using standard nomenclature, creating a silent link is easy, but more often than not, people do not name these elements and thus a name has to be created for them. Without the name, a link cannot be made.
Entities don't exist in WB
- Gene names need to be approved by WB nomenclature curators. Authors should declare any new gene names before the linking scripts are run, but authors do not always follow through. One action is to link the name to the official WBGene sequence ID. This will require that the authors have explicitly stated what the gene sequence is. Linking the name to the gene should not affect what happens in acedb, the link lives within the online paper alone. If an author is not explicit, the best option is to contact the author, if there is no response within a reasonable time, no link will be created. This situation is not a frequent event.
- Variations can be verified through the variation nameserver and a silent link made to the object. Alert the variation curator that the object is being created through the GSA JFP for that paper. Details for the variation can be entered through the form or through the Phenotype OA.
- Other entities- check with the corresponding curator as to the feasibility to create a silent link.
Formatting problems break recognition
Entity recognition relies on exact string matches and XML formatting cues. If entities are not formatted correctly in the XML then the entity may not be recognized and linked. QCFast does not allow links to be made that do not follow XML formatting restrictions, to link these entities, it is necessary to download the html, manually add the link, then upload the fixed html.
- Case (a): <i>Cbr-dpy-1</i> instead of Cbr-<i>dpy-1</i>
- Otherwise the <i> tag becomes part of the string and string matching fails on Cbr-dpy-1, even if we have it in the wormbase entity list.
- Case (b): <i>unc-119</i> instead of <i>unc</i>-<i>119</i>
- Case (c): Parentheses shouldn't be italicized since they are used as an indicator of string starts and stops for entity recognition.
- <i>cbr-unc-119</i>(<i>st20000</i>) instead of <i>cbr-unc-119(st20000)</i>
Combined entity expressions
- double mutants, double alleles
These cases also need to be dealt with by manually making changes to the XML.
Dealing with ambiguities
- We've come across entities that get mistaken for entities of a different class, this was noticed for a couple of strain names being identical to clone names.
- Jargon terminology: EMS (embryonic cell) vs EMS (mutagen)
In these cases the term is added to an exclusion from linking list. It would be ideal to try to use some sort of context specific rules to try to make these terms less ambiguous. For now, we can manually identify and link these terms when necessary.
Person names are difficult to recognize and map to a WBPersonID in an automated fashion, because of the many variations in how an author lists their name, in addition, there are cases when more than one person shares the name, initials, etc.. We will deal with this by designing a URL that will first query for a person name in the WBPerson class, that matches the extracted author name, if one is not found then the URL will point to an author page. Since all authors will get a page eventually, that link is guaranteed to be alive at some point. Q. how often do authors get linked with WBPersonID's?
We anticipate many cases where the name of a gene (or gene name) will be used across different species, making the identification of that gene problematic, not only in a single-species paper (e.g., MEX-1 exists as a name for a mammalian gene as well as for an elegans protein). We will start to deal with this by identifying all similar gene names, including synonyms, between the different organisms (SGD, flybase, mouse)