Semantic Web Applications and Tools for Life Sciences – Morning Session
November 28, 2008 2 Comments
I am currently at a meeting in Edinburgh with the title “Semantic Web Applications and Tools for Life Sciences“. The title is programmatic and it promises to be a hugely exciting meeting. As far as I can tell, the British ontological aristocracy is here and a few more besides. The following are some notes I made during the meeting.
1. Keynote: Semantic Web Technology in Translational Cancer Research (M. Krauthammer, Yale Univ.)
How to integrate semantic web technologies with the Cancer Biomedical Informatics Grid (caBIG)?
Use case: melanoma…worked on at 5 NCI sites in US: Harvard, Penn, Yale, Anderson….can measure all kinases involved in disease pathways…use semantic technologies to share and integrate data from all sites and link to other data sources…e.g. drug screening results etc…..
MelaGrid consortium: data sharing, omics integration, workflow integration for clinical trials
Data sharing: create community wide resources – a federated repository of melanoma specimens
currently caBIG uses ISO/IEC 11179 metadata standards to register CDEs (common data element) and additional annotation via NCI thesaurus concepts: example of use: caTissue…tissue tracking software (multisite banking, form definition, temporal searches etc.)
omics integration: caBIG domain models are in essence ontologies…..translate into OWL models and integrate with other ontologies (e.g. sequence ontology etc.) to align data from various sources
using Sesame as a triple store, but have performance problems….use SPARQL as query language rather than caBIGs own query language
2. Semantic Data Integration for Francisella tubularis novicida Proteomic and Genomic Data (Nadia Anwar et al.)
Why is data integration important in biology?
datainformatics in bioinformatics is nor a solved problem…there are no technologies which satisfy all the problems biologists are likely to ask, also issues with data accesss and permissions…..yet another problem is heterogeneous nature of data: information discovery is not integrated…all technologies have strengths and weaknesses…data relates – but it doesn’t overlap
Solution: semantic data integration across omes data silos….
Case Study: Francisella tularensis (bacterium, infection through airways…infects immune system….francisella can bypass macrophages….forms phagosome, but can escape from it…bioterrorism fears…..”Hittite plague” been associated with Tularemia)
available datasources: genome data…from international database….convert to simple rdf data, kegg, ncbi, GO, Poson, transcriptomics data
used data from proteomics experiment to integrate with the constructed graphs….could show that it was easy to query the whole graph…..but issues with modeling of the data and the resulting rdf graph…so some careful data modeling is still necessary….some performancce issues with datasets cotaining many reified statements…..memory problems…
Summary: In principle it’s easy – in practice it is still hard work
Use of shared lexical resources for efficient ontological engineering (Antonio Jimeno et al.)
Motivation: Health-e-Child Project (creation of an integrated (grid-based) healthcare platform for European Paediatrics
Use Case: Juvenile Rheumathoid Arthritis Ontology construction
reuse existing ontologies – Galen, NCI but….problem with alignment becuase of missing information that could facilitate mapping, also many mapping tools based on statistics….thus trust
A common terminological resource for life sciences….generate a reference thesaurus that Galen,, NCI, JRAO thesaurus to normalise term concepts
Def Thesaurus: Collection of entity names in domain with synonyms, taxonomy of more general and specific terms (DAG)…..no axiomatisation
Problems in thesaurus construction: ambiguity (retinoblastoma – gene or disease), inappropriate term labels, maintenance: thesaurus and ontologies need to be updated simultaneously now…
KASBi: Knowledge Bases Analysis in Systems Biology ()
Problem: Combining data from different data sources – use semweb rather than standard data integration systems for integration…in particualar use reasoners….
In KASBi try and integrate reasoners/semweb with traditional database tech: use semtech to generate a “query plan” which specifies how queries need to be carried out across resources
goWeb – Semantic Search Engine for the Life Science Web (Heiko Dietze)
Typical question: “What is the diagnosis for the symptoms for multiple spinal tumors and skin tumors?”, “Which organisms is FGF8 studied in?”
goWeb combines simple key-word web searching, text mining and ontologies for question answering
Keyword search in goWeb is sent to yahoo, which returns snippets. These are subsequently pushed through NLP to extract concepts and mark them up with ontology concepts…….use ontolgies to further filter results…..
Path Explorer: Service Mining for Biological Pathways on the Web (George Zheng)
Two major biological data representation approaches: free text(discoverable but not invocable), computer models (constructed but made available in isolated environment – invocable but not discoverable)
Solution: model biological processes using web service operations (aim: to be invocable and discoverable) pathways of service oriented processes canbe discovered and invoked
SOA: service providers publish services into registry where they can be discovered by service providers
DAMN – slides are much to small…can’t see anything….”entities are service providers and service consumers”
….ook…..he’s lost me now – I can’t see anything anymore…..
Close integration of ML and NLP tools in …
Scope: Fine grained semantic annotation: eg he GenE protein inhibits……mark up GenE protein as a protein, inhibits as a negative interaction etc…..
Availability of NLP Pipeline….Alvis/A3P, GATE, UMA but domain specific NLP resources are rare
focus on target knowledge ensures learnability
rigorous manual annotation
high quality annotation and low vvlumes require proper nrmalisation of training corpora (syntactic dependencies vs shallow clues)
clarification of different annotatoon tasks and knowledge – consistency between NE ype and semantics
Fine grained annotation is feasible and necessary for high quality services: i.e. in verticals and science….
Right – time for lunch and a break. I have only captured aspects of the presentations and stuff that resonated with me at the time….so please nobody shoot me if they think I haven’t grabbed the most fundamental points….Link to the slides from the event is here