Re: Notes from informal Demo F2F

Hi Eric,

I was the ?. I attended for about 1.5 hours of the call and then I had 
to leave quietly (I didn't want to disturb the discussion) for teaching 
a class. I think the suggestions are great. It's important to show the 
benefits that SW can potentially bring to the healthcare and life 
science research community. Some of these benefits, as you and others 
mentioned, include data aggregation, integration, and reasoning over a 
variety of data sets. It it also important to bear in the mind the 
intended audience. As people are already aware, most folks at the WWW 
conference are technologists so we may need to keep the biology part as 
readable to them as possible. These folks may also want to see how HCLS 
can help drive SW research and development (just like CAD/CAM was one of 
the main driving forces for object oriented modeling research). They may 
like to see what challenges certain types of biomedical data could bring 
to the SW research community.

Just as an exercise, based on the AD/PD use case, I've been 
contemplating how to model Abeta which is a kind of peptide. For 
example, I can define a class called PEPTIDE. One of its properties can 
be amino_acid_sequence_Length (or aa_seq_len). My question is that can 
we use some OWL constructs to create different ABETA classes based on 
the PEPTIDE class? For example, ABETA40 and ABETA42 may be defined as 
PEPTIDE with aa_seq_len=40 and PEPTIDE with aa_seq_len=42, respectively. 
I was told that this type of class modeling may not be possible in OWL. 
If this is true,what should we do? Again, I'm neither an OWL ontology 
expert nor a neuroscience expert. I'm just who I'm. This is just my 
2-cent of the 2-cent thought.

Best,

-Kei

PS: We will make the RDF version of NeuronDB and ModelDB available as 
soon as possible for the demo group to test and use (we are in the final 
process of data translation and debugging).

Eric Neumann wrote:

>
> At one point during the informal F2F, the folks on the call (Karen 
> Skinner, Eric P, John Barkely, Scott Marshall, Eric N, ?) were cut off 
> from the conference, but we ended up having a discussion amonst 
> ourselves, where we came up with a few suggestions regarding the demo:
>
> - it might be compelling to use enough data (already known facts) to 
> help make some interesting associations (Scott M's comment of 
> 'manufactured serendipity'), which could then be annotated and used to 
> propose hypotheses.
> - use existing and familiar sources of genes and diseases (e.g., 
> Entrez-Gene, Uniprot, etc, around AD/PD) for deeper data aggregation 
> to show how RDF and SPARQL (this may already be what some of you are 
> planning on doing)
> - As part of the scernario using the known aggregate of facts, add a 
> few *select* hypotheses (triple graphs), that would make major 
> connections with the rest of the graph that would function as a 
> "bridge" across the data and models; Show the new insights from this 
> merged compositeby re-applying queries that now retireve more 
> connections. One example Karen had was around the MPTP/MPP+ mechanism 
> for some forms of PD.
>
> The latter point will probably need to be "manufactured", since we 
> cannot rely on trying to uncover real scientific insights in time for 
> this demo-- would be nice though!
>
> It was also apparent that we need to identify what the specific "aha" 
> or "cool factor" is in the demo when shown to the select audience; it 
> appears to me taht it should be about the RDF or OWL nature of the 
> data, federated SPARQL querying, and ability to add new data or 
> annotations and get even more out of the construct.
>
> -- 2¢ from those who called in.
> Eric
>
>
> -----Original Message-----
> From: public-semweb-lifesci-request@w3.org on behalf of Alan Ruttenberg
> Sent: Mon 3/5/2007 10:35 PM
> To: public-semweb-lifesci@w3.org
> Subject: Notes from informal Demo F2F
>
>
> Summary:
>
> We reviewed the use case, and then, starting in one area of the use 
> case - the connection between ADDL and impairment of long term 
> potentiation deficits - highlighted on page 2 of the updated use case 
> [1] - and started to trace out a series of steps that lead towards 
> determining relevant sets of gene  and then to images of expression 
> of those genes. Along the way we reviewed web pages and databases to 
> verify what information was available. We then collected a list of on 
> line databases that would enable
> the kinds of queries we were exploring.
>
> [1] http://esw.w3.org/topic/HCLSIG_BioRDF_Subgroup/Demo_Thoughts/
> Alzheimer%27s_Proposal?
> action=AttachFile&do=get&target=AD_PDUseCase_02-26-07.pdf
>
> Start:
>
> Hypothesis+Abeta -> ADDL (SWAN. Note that these areas of SWAN are not 
> yet populated but June and Gwen are working on this).
>
> We attempted to linking ABeta to LTP, but this did not find the 
> requisite links. What we did was:
>
> Go to http://senselab.med.yale.edu/senselab/modeldb/. Search for 
> "ltp" (could also search for "long term potentiation"
> This is a text search, and the matches are typically against the 
> titles of citations.
> Each result links to a model, which lists some cell types, some 
> receptor families, some transmitters.
>
> We reviewed the receptors and searched for them in NeuronDB but did 
> not find links to processes such as long term potentiation.
>
> --- Starting over
>
> Go to http://senselab.med.yale.edu/BrainPharm/eavData.asp?
> db=1&c=122&o=5904
> Links from Alzheimer's disease to  CA1 Pyramidal Neuron
>
> Click on CA1 Pyramidal Neuron
> Link to http://senselab.med.yale.edu/BrainPharm/NeuronDB/
> ndbEavSum.asp?id=5588&mo=4&re=
>
> CA1 Pyramidal Neuron Extracellular Elements : Abeta (current I A)
> CA1 Pyramidal Neuron Intrinsic Currents : I A. Click on I A.
>
> http://senselab.med.yale.edu/BrainPharm/eavData.asp?o=5545
> Links to K+
>
> Need to know that current of K+ is carried by Potassium Channels.
>
> However, even though some text indicates LTP, there is no explicit 
> link. We do have a link from neurons of interest
> to genes however, if we are able to look up proteins associated with 
> potassium channel function.
> Browse GO we do find these associations.
>
> ---- Approach using LTP as a key from a different direction.
>
> There is a mesh term
> http://www.nlm.nih.gov/cgi/mesh/2007/MB_cgi?mode=&term=Long-Term 
> <http://www.nlm.nih.gov/cgi/mesh/2007/MB_cgi?mode=&term=Long-Term>
> +Potentiation&field=entry
>
> We can link to pubmed papers using mesh term headings.
>
> http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?
> CMD=search&DB=pubmed&term=Long-Term+Potentiation%5Bmh%5D&cmd
>
> There is also Potassium Channel
> http://www.nlm.nih.gov/cgi/mesh/2007/MB_cgi?mode=&term=Potassium 
> <http://www.nlm.nih.gov/cgi/mesh/2007/MB_cgi?mode=&term=Potassium>
> +Channels&field=entry
> and Pyramidal cells
> http://www.nlm.nih.gov/cgi/mesh/2007/MB_cgi?mode=&term=Pyramidal 
> <http://www.nlm.nih.gov/cgi/mesh/2007/MB_cgi?mode=&term=Pyramidal>
> +Cells&field=entry
>
> So if we have a Pubmed -> Mesh mapping, we are able to use this to 
> navigate from mesh term to mesh term,
> for example, from LTP to cell types or to channel types.
>
> This also requires that we have the MESH tree encoded. Note, however, 
> that this MESH->MESH links are potentially unreliable
> as they could be the product of unrelated discussions in the paper 
> used to do the mapping.
>
> --- From pubmed to genes.
>
> We can also go from pubmed to gene. This is seen in the user 
> interface via the "links" button on a pubmed abstract page.
> e.g.
> http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?
> db=pubmed&cmd=Retrieve&dopt=AbstractPlus&list_uids=17251421&query_hl=5&i
> tool=pubmed_docsum
>
> Choose links, then gene, then you get:
>
> http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?
> itool=pubmed_AbstractPlus&db=pubmed&cmd=Display&dopt=pubmed_gene&from_ui
> d=17251421
>
> The mapping of pubmed -> gene is not in the pubmed records, but is 
> kept on the gene side.
> There are entries in the Entrez gene asn or xml record, as well as 
> the generifs, which associate a gene with a paper.
>
> Along with the Gene ontology, we now have two ways to navigate from 
> "Potassium channel" to gene.
>
> -- From gene to brain region
>
> We can get BrainRegion from gene in a number of ways:
>
> Using expression data: The Allen Brain Map has, for a select group of 
> regions and structures, lists of genes
> expressed in those structures. Some of these are at:
>
> http://community.brain-map.org/confluence/display/DataAnno/Home
>
> Others can be scraped from the results of queries at http://brain-
> map.org/welcome.do (click on the "Anatomic Search" tab)
> (Alan has these)
>
> Similarly, Gensat (ftp://ftp.ncbi.nih.gov/RawData/
> GENSAT-20050125.xml.gz) Has images associated with genes
> and annotations that say which areas/cell types have which patterns 
> of expressions of those genes.
> (Alan has these)
>
> --- From Cell type to brain region.
>
> In addition to the mappings implicit in Gensat,
> BAMS  http://brancusi.usc.edu/bkms/xml/swanson-98.xml has mappings of 
> cell types to brain regions
>
> -- Another route for getting Alzheimer related genes:
>
> We can also get from Alzheimer to gene via Alzgene
> There is a popup of genes on http://www.alzforum.org/res/com/gen/
> alzgene/default.asp, which records
> studies which find associations between mutations of these genes with 
> Alzheimer disease.
>
> --- Some other resources
>
> We will want human to mouse gene mappings for better navigating from 
> pubmed.
> Available at:  http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?
> db=homologene
>
> For linking gene to protein and protein products such as cleaved 
> proteins, June has started curating these from various sources.
> Uniprot has such mapping, it is only textual. See, e.g.,  http://
> www.pir.uniprot.org/cgi-bin/upEntry?id=P05067. Look for the 
> "contains" field.
>
> Another resource for relating phenotype, process to genes is the JAX 
> Mammalian Phenotype ontology/
> Associated with each phenotype, are the alleles/genes associated with 
> the phenotype.
>
> http://www.informatics.jax.org/searches/MP_form.shtml
> For example, abnormal hippocampus function http://
> www.informatics.jax.org/javawi2/servlet/WIFetch?
> page=mpAnnotSummary&id=MP:0001895
>
> Gwen and Don will review the terms to see which would be relevant to 
> processes associated with Alzheimers, and this gives us another
> route to narrow/expand genes of interest.
>
> Summarizing data sources which we will use for the next step of 
> exploration. Some we have, some we need to acquire/convert to RDF.
>
> Mesh
> Pubmed -> Mesh Term
> GO -> Associations (Genes)
> Entrez Gene -> Pubmed
> Entrez GeneRif (= gene->Pubmed)
> NeuronDB/Brainpharm
> Homologene
> Alzgene (Alzheimer Disease -> Associated Gene)
> JAX: Mouse Phenotpe -> Gene
> Uniprot Gene-> splice form, gene product.
> BAMS swanson-98.xml
> Allen/Gensat gene->image, gene-> regions/expressed
>
> Alan's comment: Process for conversion to RDF should be 2 step
> 1) Propose model, give sample entry for review
> 2) Upon agreement on model translate rests/load into triple store for 
> query.
>
> Action: Candidates for doing conversions: Alan, EricP, Matthias, Don
> (?) to talk and split tasks.
> Action: Don and Gwen to review phenotypes
> Action: June to curate trying to write pseudo-triples to make 
> conversion to real triples easier.
>
> -Alan
>
>
>
>
>
>
>
>

Received on Wednesday, 7 March 2007 03:35:08 UTC