Transcript Data provenance in biomedical discovery

Data provenance in
biomedical discovery
Donald Dunbar
Queen’s Medical Research Institute
University of Edinburgh
Workshop on Principles of Provenance in Databases
May 21st 2008
Background
biomedical research
basic & clinical science
animal, cell models, patients
genes, proteins, pathways
data analysis & mining
publication
Biomedical discovery
• Looking for contribution to
– human health and disease
• In house experiments
– data workflows
– knowledge capture
• Use public databases
– many data types
– integration is a problem
Databases we use
sequence
structure
function
expression
domain specific
Data workflows
database
experiment 1
experiment 2 raw data
spreadsheet
publication
processed
data
database
calculations
Data workflows
IN
copy and paste
open from file
‘algorithm’
copy and paste
BUT:
web services
automated tools & databases
bioinformatics workflows
save to file
OUT
Bioinformatics workflows
Is our field changing?
databases
experiments
knowledge
knowledgebase
Knowledge capture
Knowledge capture
What provenance to we need?
Example:
Gene expression in a transgenic animal
gene expression measurements
gene annotation
where, when
public databases
which identifiers
integration
when, what, how
output from machine
how
processing
what and how did we select genes
data mining
…
What provenance to we need?
Example:
Curated protein database
database links
expert data
contributor, date
curator input
source, identifiers, dates
development
verify, add, delete, modify
archive
versions, dates
Curated database
schema & interface changes
What do we do now (for provenance)?
• We trust the main data providers a lot!
– a pragmatic approach
• We use tools and note the settings
– rarely fully
• We put extra fields in our databases
– source, modify date
• We deposit our data in public repositories
– but only when we need to
What might we do next?
• Use workflow tools like Taverna
– capture workflow provenance
• Build provenance tool & database
– widely applicable
• Make provenance more visible to biologists
– so they value and use it
Conclusions
•
•
•
•
•
In biology we don’t do provenance well (yet)
We use databases and manual workflows
We implement rudimentary provenance
We should build useful provenance tools
We need to make provenance visible