web Semantic methodologies for statistical genomics VJ Carey Banff 2004
!
"
#
Overview
motivations: needs and benets of explicit semantics examples: reading a journal with R extracting design ontology from MAGE-OM PPI graph from EBI-Intact Formalism: Graphical models for general infosets, with illustration on a knockout database Inference (model entailments, rule entailments) Rswub: R interfaces to Jena (HP) and LSID (IBM) Rself: Self-analyzing bioinformatic data structures
!
"
Caveats
semantic web initiatives very ambitious, but cant provide a universal solution for self-documenting data resources meaning an elusive concept, never exhausted, can depend on acceptance of interpretive authority best hopes: reduce the costs of coupling interpretive resources to analytic resources reduce the costs of making software that protably uses interpretive resources it would be appropriate to improve the programmatic interpretability of statistical databases generally...
! "
Technical caveat
Information models (RDF, OWL, LSID) are W3C/OMG specications, evolving over time Applicable toolsets have various APIs Ive dened interfaces to these to allow R users to explore the technologies Some functionality is missing, but emphasis is placed on exploiting reectance (thanks to Duncan Temple Lang) we can learn about the underlying API by interrogating the objects it produces we can use aspects of this API even though I did not explicitly dene an interface, using the reference
! "
Motivation: Multispecies co-expression study Stuart et al, Science 10 Oct 2003, p249 gene co-expression network/conserved genetic module discovery metagenes in 4 genomes identied on basis of sequence similarity co-expression between metagenes assessed by correlations across 2700+ microarrays metagenes situated in the plane using a proprietary layout for proximity graph derived from expression-correlation-signicance distance clusters in the plane interrogated for shared GO function
! "
Stuart et al base resource
pairs of 6307 metagenes determined by evaluating correlation within organisms across 2700+ microarrays
!
"
Metagene example
!
"
Stuart et al nal product
expression correlation pvals across expts determine pairwise distances; proprietary layout; spatial density estimation
!
"
It would be nice if we could
amalgamate independently archived experimental results with minimal prospective handshaking have explicit access to interpretive resources annotation evidence codes (often ignored in GO utilities) experimental protocols biological annotation in programmatic format algorithm specications algorithm implementations make extended use of this product (new data or new interpretive methods) with minimal software
!
"
Semantic web can help
amalgamation of independent data resources: protocols and tools for harmonization and combination of RDF/OWL models access to interpretive resources:...
carey