Nesc invited presentation: Semantic Provenance and Linked Open Data

“looking back
to gaze forwards”
Janus
Provenance

LOP: Linked Open Provenance
(was: from Workﬂ
ows to Semantic Provenance
and Linked Open Data)
Paolo Missier Jun Zhao Satya S. Sahoo
Newcastle University Amit Sheth
(University of Manchester) University of Oxford, UK Wright State University, USA

Nesc workshop:
Understanding Provenance and Linked Open Data

Edinburgh, March 30th, 2011

Baseline provenance of a workflow run
QTL → mmu:12575
Ensembl Genes
v1 ... vn w

Ensembl Gene → Ensembl Gene →
Uniprot Gene Entrez Gene
path:mmu04012
exec
Uniprot Gene → Entrez Gene → a1 ... an b1 ... bm
Kegg Gene Kegg Gene
mmu:26416

merge gene IDs

path:mmu04010 y11 ymn
Gene → Pathway
...

path:mmu04010→derives_from→mmu:26416
path:mmu04012→derives_from→mmu:12575

• The graph encodes all direct data dependency relations
• Baseline query model: compute paths amongst sets of nodes
• Transitive closure over data dependency relations

2
P.Missier - Provenance and Linked Data workshop
Edinburgh March 2011

Key ideas
• Janus:
– a semantic provenance model with domain-specific
extensions
– designed around the Taverna workflow model

• From domain-agnostic provenance graphs
• To domain-aware graphs through explicit
annotations

• From local provenance graphs and queries
scoped to the graph
• To
– Graphs published as Linked Data
– Queries extended into the Web of Data

3

Reference user questions

QTL →
Ensembl Genes Q0: Find all intermediate and initial input
values that contribute to the computation of a
certain output value.

Q1. Find all those genes within the input QTL
Uniprot Gene → Entrez Gene →
region that are involved in a given KEGG
Kegg Gene Kegg Gene pathway.
merge gene IDs
Q2: Find all Uniprot-sourced genes

Gene → Pathway

Q3: Find all Entrez genes that encode proteins involved in ATP binding (go:
0005524).

Q4: List relevant PubMed publications for the pathways listed in the result set.

4

Query types and model capabilities

Query formulation effort Annotation requirements Query Scope

Q0 - Requires knowledge of
process structure and
data values
No annotations required Single run graph or
Multi-run graphs
- Graphical query
constructor may be
available
Q1 Q2
Requires domain annotations
Use of domain terms Single run graph or
on workflow tasks and on data
facilitates query formulation Multi-run graphs
values

Q3 Q4
- Use of domain terms - Requires domain annotations
facilitates query on workflow tasks and on
formulation. data values
The Web of Data
- Can be integrated with - Relies on completeness of
browsers for LoD sources Linked Data Sources

5

Janus structure

6

Janus structure
port
v1 v2 v3 value

X1 X2 X3 X1 X2 X3
port
P exec P_inst
Y1 Y2 Y1 Y2
processor
processor exec
spec w1 w2

6

Annotated provenance

Annotated workflow Annotated provenance graph

QTL → Gene
Ensembl Genes


exec ...
Kegg Gene Kegg Gene

merge gene IDs

Kegg
Gene → Pathway

Q1 Q2

7

Desiderata: from structure to values annotations

8


protein
sequence

X1 X2 X3
interpro
P match
report
Y1 Y2

processor
interpro spec
scan

8


protein
sequence

X1 X2 X3
interpro
P match
report
Y1 Y2

processor
interpro spec
scan

protein
sequence
port
v1 v2 v3 value

X1
exec
X2 X3
port
P_inst

interpro Y1 Y2
processor interpro
scan exec
w1 w2 match
8 report

Annotations propagation rules
X rdf:type Port C = {c} X has value type c
X has value v v rdf:type PortValue
v rdf:type C denotes
data type
in the PL
protein sense
has_value_type sequence

X1 X2 X3
interpro
P match
report
Y1 Y2

processor
interpro spec
scan

9

data type
in the PL
protein sense

X1 X2 X3
interpro
P match
report
Y1 Y2

processor
?
interpro spec port
scan v1 v2 v3 value

X1 X2 X3
port
P_inst

interpro Y1 Y2
processor interpro
scan exec
w1 w2 match
9 report

data type
in the PL
protein sense

X1 X2 X3
interpro
P match
report
Y1 Y2
protein
processor sequence
interpro spec port
scan v1 v2 v3 value

X1 X2 X3
port
P_inst

interpro Y1 Y2
processor interpro
scan exec
w1 w2 match
9 report

Annotations as semantic overlay

v1 w1

X3

Y2
Provenance graph has_port_value has_port_value

X2

P
fragment

Y1
X1
vn wm

has-input-type Pathway has-output-type
search
service

instance-of
instance-of instance-of

Gene v1 w1 Pathway
has-source
X3

has-source
has_port_value Y2 has_port_value
X2

P

instance-of
instance-of
Y1

Kegg wm
X1

vn Kegg
has-source has-source

10

Extensions to Linked Data


QTL →
Ensembl Genes

exec ...
Kegg Gene Kegg Gene

merge gene IDs

Gene → Pathway

11

Extensions to Linked Data


QTL →
Ensembl Genes

exec ...
Kegg Gene Kegg Gene

merge gene IDs

Gene → Pathway

- Publish
- I - Map IDs
- II - query

11

I - Mapping data values to LoD URIs
In our prototype we map data values to Bio2RDF as follows:
Entrez Genes

Uniprot Genes

KEGG Genes

KEGG Pathways

<rdf:Description rdf:about="http://purl.org/net/taverna/janus/create_report/entrezGeneId">
<janus:has_value_binding rdf:resource="http://purl.org/net/taverna/janus/test18"/>

<rdf:Description rdf:about="http://purl.org/net/taverna/janus/test18">
<rdf:type rdf:resource="http://purl.org/net/taverna/janus#port_value"/>
<rdfs:comment>11835</rdfs:comment>
<rdf:type rdf:resource="http://purl.org/obo/owl/sequence#gene"/>
<janus:has_source rdf:resource="http://purl.org/net/taverna/janus#entrez_gene"/>

<rdf:Description rdf:about="http://purl.org/net/taverna/janus/test18">
<rdfs:seeAlso rdf:resource="http://bio2rdf.org/geneid:11835"/>
12

II - extended queries in the LoD setting
Strategy:
- use the SQUIN LoD query engine to query multiple “Web of Data” sources
- only Bio2RDF in our case
- combine graph patterns on local provenance with conditions on remote LoD
graphs

Q5: Find all Entrez genes that encode proteins involved in ATP binding
(GO:0005524).

PREFIX uniprot: <http://purl.uniprot.org/core/>
PREFIX : <http://www.taverna.org.uk/janus#>
SELECT distinct ?entrezgene
WHERE {
?protein uniprot:classifiedWith <http://bio2rdf.org/go:0005524> . Bio2RDF
?entrezgene <http://bio2rdf.org/bio2rdf_resource:xPath> ?protein .
?gene rdfs:seeAlso ?entrezgene
?gene rdf:type :port_gene
?gene :has_source :entrez_gene . }
local provenance graph

13

Current status
Current Taverna provenance architecture:
Lab prototype Production
– “Export as...” Janus RDF – “native” (relational) graphs
– currently only queried using – simple, efficient query language on
SPARQL native provenance
e
– manually published
– manually annotated
Taverna
Taverna <<Events stream>> Provenance
runtime events capture

query <scope ... /> Lineage
<select .../> relational
query
<focus .../>
processor DB
query response
OPM graph

complete Janus RDF
graph exporter

Provenance
API (Java)
14

Summary and References
• Janus: a semantic model for workflow provenance
– OWL ontology, extension of Provenir
– should include attribution + system level provenance
– alignment with OPM?

• Domain-aware graphs through annotations:
– automatically propagated from workflow annotations when possible
– but in practice no real workflows are annotated

• LoD integration:
– powerful provenance publishing and query broadening
– mapping rules currently limited
– no completeness guarantee -- all joins are outer joins!

P. Missier, S.S. Sahoo, J. Zhao, A. Sheth, and C. Goble, "Janus: from Workflows to Semantic
Provenance and Linked Open Data," Procs. IPAW 2010, Troy, NY: 2010

S.S. Sahoo, J. Zhao, and P. Missier, "Extending Semantic provenance into the Web of Data,"
Internet Computing, special issue on Provenance in Web Applications, vol. to appear, 2011.
15

Nesc invited presentation: Semantic Provenance and Linked Open Data

Recomendados

Recomendados

Mais conteúdo relacionado

Destaque

Destaque (10)

Semelhante a Nesc invited presentation: Semantic Provenance and Linked Open Data

Semelhante a Nesc invited presentation: Semantic Provenance and Linked Open Data (20)

Mais de Paolo Missier

Mais de Paolo Missier (20)

Nesc invited presentation: Semantic Provenance and Linked Open Data

Notas do Editor