Adaptive Information Disclosure (AID)

CWA prototype - Calpain 3 (continued)

Marco — Tue, 05 May 2009 20:45:43 +0000

Update from latest telcon by Eric:

The knowledge base triples do not contain human readable labels (yet); these could come from UMLS, but the UMLS upload was not yet successful (it contains a limited set of triples.
Another approach may be to try to create REST calls to public databases (i.e. URIs).
The cwa repository should contain lovd data now.

I will try to reach LOVD data, and perhaps some other data to link out from.

First I had to figure out how to do regex on parts of URIs; there is a string representation function; this works:
PREFIX cwa: PREFIX rdf: SELECT ?cwa_concept ?relation ?cwa_concept2 WHERE { ?cwa_concept ?relation ?cwa_concept2. FILTER (regex(str(?cwa_concept),”C3000052″,”i”)) }

This filter
FILTER (regex(str(?cwa_concept),”lovd”,”i”) || regex(str(?relation),”lovd”,”i”) || regex(str(?cwa_concept2),”lovd”,”i”))
in the query above did not give any results; I’ve asked Erik what I should do to get something from lovd.

CWA prototype - Calpain 3 (continued)

Marco — Tue, 05 May 2009 08:49:11 +0000

I have difficulty finding a path to more information about Calpain-3 via the umls identifier C3000052. This queryPREFIX umls: SELECT ?umls_concept ?umls_relation ?umls_concept2 WHERE { ?umls_concept ?umls_relation ?umls_concept2. FILTER (?umls_concept = umls:C3000052) }
has no results, while filtering on umls:A8400986 does (used this id just as a test).

Because of this and that the response time for broad searches on umls on the cloud server (e.g. with regex) is slow, I want to have a look at mesh. The mesh subset is also on our aida server. The mesh id for human Calpain-3 on the nlm mesh website appears to be C105884.

I tried this query on the cloud umls repository:PREFIX umls: SELECT ?umls_concept ?umls_relation ?umls_concept2 WHERE { ?umls_concept ?umls_relation ?umls_concept2. FILTER (?umls_concept = || ?umls_concept2 = ) }
gave 36 triples with interesting information, but not a name yet (which is what I am looking for). But it does provide a refseq and a link to a document. However, I have not found this via a link with C3000052 of course.

Suggestion

All in all, I have stumbled upon the well known bioinformatics problem of identifiers, which leads me to propose a suggestion for CWA. Although I appreciate any attempt to find unambiguous ’standard’ identifiers, I think experience tells us that most attempts have just added a new types of identifiers. Therefore, I suggest that we focus on reducing missing links (which seems to be my problem here). This will not be a perfect solution, but perhaps a good one I imagine we facilitate a community approach, based on these principles:

any link to any of the many ’standard’ identifiers links in with a network of interlinked identifiers such that we can reach the identifier we need
any link created by a user adds to the network, thereby reducing missing links (hence, the community is making the network)
tools can be built on this network:
1. e.g. a path traverser for when the required information is a number of triples away along a path. (e.g I was looking for a human readable label); gradually we could build in intelligence, such as SKOS label discovery.
2. A web service for identifier discovery based on this would be really useful in workflows. Such a service would hide the RDF details from the queries (you could just say: ‘umls’, or ‘mesh’ or ‘refseq’ for the output type; I imagine the traversed path is a second output for validation)
3. Would this be a connection with Okkam (Staphano Bocconi)? I think this is doable, useful for users, and enough for a computer scientist to produce papers on (imho I don’t think Okkam can produce something useful on both sides unless a community is enabled to do most of the work).

SeRQL to SPARQL for CWA query

Marco — Mon, 04 May 2009 12:24:38 +0000

I tried to convert a serql query to a sparql query, but I could not find a proper equivalent of the Localname (and Namespace) functions in SeRQL, or documentation/examples of how to do this with FILTER functionsÂ were too difficult to find. The solution chosen for SPARQL seems more powerful and extensible, but it is definitely more complicated and could use more examples (or easier to find with Google).

SeRQL:

SELECT DISTINCT cwa_concept, relation, cwa_concept2 FROM {cwa_concept} relation {cwa_concept2} WHERE Localname(cwa_concept)= "C3000052" USING NAMESPACE cwa = , rdf =
My SPARQL ‘almost’ equivalent:
PREFIX cwa: PREFIX rdf: SELECT ?cwa_concept ?relation ?cwa_concept2 WHERE { ?cwa_concept ?relation ?cwa_concept2. FILTER (?cwa_concept = ) }

CWA prototype - Calpain 3

Marco — Sun, 03 May 2009 15:32:27 +0000

Uniprot contains several human versions (disregarding entries described as fragments or ‘highly similar to’)

P20807
Q9BQC8
Q496N2
Q762C8

I will concentrate on http://www.uniprot.org/uniprot/P20807

However, Erik’s CWA servlet (http://aneurist.erasmusmc.nl/CWA/servlet/pop) tells me the UMLS identifier is umls/C3000052. I think that is used as the identifier in the CWA knowledge base, so I will continue with that.

The query

SELECT DISTINCT cwa_concept, relation, cwa_concept2
FROM
{cwa_concept} relation {cwa_concept2}
WHERE
Localname(cwa_concept)= "C3000052"
USING NAMESPACE
cwa = ,
rdf =

gives all relations of the UMLS term identifyer. ‘relation’ appears to be one of cwa:related, cwa:relation, rdf:instanceOf; there appears to be no further information (e.g. human readable labels) in this repository.

The SPARQL ‘amost’ equivalent (couldn’t find ‘Localname()’ equivalent: see this post):
PREFIX cwa: PREFIX rdf: SELECT ?cwa_concept ?relation ?cwa_concept2 WHERE { ?cwa_concept ?relation ?cwa_concept2. FILTER (?cwa_concept = ) }
[CHECK UMLS repository]

CWA prototype triples exploration

Marco — Sun, 03 May 2009 15:30:53 +0000

Exploring one relation:

http://ec2.homelinux.org:8080/openrdf-workbench/repositories/cwa/explore?resource=%3Chttp%3A%2F%2Fwww.nbic.nl%2Fcwa%2Frelation%2FC0033684%23C3041379%231240640890649%3E

Annotation:

http://www.uniprot.org/uniprot/a16_anoga links trough to Uniprot Protein A16 precursor (http://www.uniprot.org/uniprot/Q93118)

relation concepts are annotated with a uniprot reference (cwa:annotation)
This:
is a cooccurence relation (cwa:typeRelation cwa:cooccurrence)
- It is related to (cwa:relation)
- I can’t make out the type of
- cwa:relation
cwa:has_query shows the user query (I think)
cwa:related binds relation terms {concept1} cwa:related {concept2}
How do I go from a relation term to a cooccurence?

CWA prototype class and relation types

Marco — Wed, 29 Apr 2009 11:10:22 +0000

CWA classes

SELECT DISTINCT cwa_class
FROM
{cwa_instance} rdf:instanceOf {cwa_class}
USING NAMESPACE
cwa = ,
rdf =

cwa_class:

   cwa:RelationTerm

CWA relations

SELECT DISTINCT cwa_relation
FROM
{cwa_concept1} cwa_relation {cwa_concept2}
WHERE
Namespace(cwa_relation)=cwa:
USING NAMESPACE
cwa = ,
rdf =

Results in:

cwa_relation:
Â Â Â Â Â Â Â  cwa:related
Â Â Â Â Â Â Â  cwa:relation
Â Â Â Â Â Â Â  cwa:typeRelation
Â Â Â Â Â Â Â  cwa:strength
Â Â Â Â Â Â Â  cwa:has_queryÂ Â 
Â Â Â Â Â Â Â  cwa:discovered_by
Â Â Â Â Â Â Â  cwa:timestamp
Â Â Â Â Â Â Â  cwa:annotation

CWA relation types

SELECT DISTINCT cwa_type
FROM
{cwa_concept} cwa:typeRelation {cwa_type}
USING NAMESPACE
cwa = ,
rdf =

cwa_type

        cwa:cooccurrence

Protected: Apr 10 Meeting Minutes

AID — Mon, 28 Apr 2008 11:13:12 +0000

Taverna workflow enactor op dev.wongiseng.homelinux.org

Marco — Thu, 10 Apr 2008 15:33:02 +0000

Tijdens mijn bezoek in Manchester hebben een aantal myGrid mensen op Scott en mijn verzoek een Taverna workflow enactor onder 9999 geinstalleerd (webapps/remotetaverna). We kunnen hiermee bv experimenteren met workflows laten draaien onder een eigen web pagina (via REST calls), maar ook om vanuit Taverna een workflow niet op je client maar op een server te draaien.

Een paar dingen:
* Tomcat-dev is een paar keer herstart, mijn excuses als iemand daar last van heeft gehad. Het was ongepland.
* Onder tomcat-dev/webapps zijn ‘remotetaverna.war’ en remotetaverna erbij gekomen
* In catalina.policy is een regel gezet om voor de database die bij remotetaverna hoort (Edgar, misschien wil je die bekijken).
* Die database staat in /home/aida/remoteTavernaDB
* Login is admin en wachtwoord is geheim, maar het liefst je eigen naam en wachtwoord aanmaken
* De runner is op adres: http://dev.wongiseng.homelinux.org/remotetaverna/v1/
* Op myExperiment heb ik alvast een ‘runner’ aangemaakt: http://www.myexperiment.org/runners/5

Complications with mesh

Marco — Sun, 30 Mar 2008 22:05:55 +0000

I would like to limit the search results for the following related reasons:

Protein names are not unique between organisms. They are closer to unique within an organism. Hence, if I could limit the search to papers e.g. in the human context, I would be more certain of the meaning of a protein name.
Martijn Scheumie’s uniprot service that I use for validation works on human uniprots only. Without limiting the documents to search, it is more likely that I will erroneously instantiate a protein with a human uniprot id, while it is e.g. a yeast protein.
More extracted proteins would be valid human proteins if I could preselect papers on organism.

mesh does not seem a solution:

mesh headings include ‘organism’, but in my very rough survey it seems this categorisation is not always used. Many papers returned by ‘EZH2′ that are at least relevant to human cell biology are not tagged with ‘Humans’ or any organism.

adding ‘human’ and synonyms to the search query (i.e. not only mesh) is not a solution

abstracts do not necessarily contain a reference to the organism in which context it is written (experts can tell or it is not so important for expert readers).

concept profiles could be useful

A concept profile might be indicative of the organism the paper is about.

Complications of extracting protein names

Marco — Sun, 30 Mar 2008 08:02:58 +0000

I tried applyCRF, experiences:

The list output combines the ‘B-Protein’ and ‘I-Protein’ of the other output formats
The resulting names are often more than a gene symbol (e.g. ‘NRG1 gene’)
Sometimes a B-Protein is truncated and a gene symbol extension is the first I-Protein (e.g. ‘DNAse’, ‘I’)
B-Proteins can have little mistakes (e.g. ‘STA1,’), with regexp I clear comma’s and points at the end of B-Proteins
Conclusion: I do everything: list output (B-Protein+I-Protein), B-Protein only (with and without [,.]) before validation

I tested applyCRF with Matijn Schuemie’s UniProt service. The purpose is to filter our false positives. I found that

Many proteins names give no result; these are generally not human (Scheumie’s service is human only)
The DDBJ service, GetGene_DDBJentry or GetProd_DDBJentry from Japan can also be used
- works on many organisms
- multiple results: from which organism is the extracted protein name?
- more unique names could be retrieved if the organism is know
Conclusion: organism is essential to identify a protein name: a gene symbol is (more or less) unique within a species. Add organism name to document search? What to do when id is not unique?

ToDo

Compare NER and applyCRF in one workflow
Consider taking up organism name in document search query.