NHM Data Portal:first steps toward the Graph-of-Life
Vince Smith, Ben Scott & Ed BakerInformatics & Digital Collections Group, NHM London
SPNHC, Berlin, 23 June 2016
NHM CollectionCollection area No of objects No of type
specimens Physical register
Digital data
Palaeontology 6,919,207 43,146 2,364,232 340,636 Mineralogy 423,563 615 425,000 402,727 Botany 5,863,000 172,750 127,200 645,222 Entomology 33,753,257 612,796 57,197 255,000 Zoology 27,501,350 325,000 1,986,000 1,160,216 Library & archives 5,460,000 - - - TOTAL 79,920,377 1,154,307 4,959,629 2,803,801
<3% of NHM specimens are digitised, & even fewer are ‘computable’
Citizen science
Big, open, linked dataHigh-throughput digitisation
Data portal and tools Text mining
Robotics
Digital Science at the NHM
Citizen science
Big, open, linked dataHigh-throughput digitisation
Data portal and tools Text mining
Robotics
Digital Science at the NHM
NHM Digital Collections Access, pre-2015• Developed with the best of intentions, but…• 23 separate interfaces• Hard to find, cite, access and integrate• No maps, few images, slow, no statistics, no export,
few updates, no authors, no citation mechanisms, no GBIF connection
NHM Data Portal• Discovery of NHM collections & research data• Easy access & reuse to promote collaboration
(website, API, R-package, RDF & direct download)• 3.7m records, >1m images (+sound, video & 3D) • Integrates with our collection management
system (weekly) & DAM system (for images)• Traffic light data quality indicators• Stable, citable (DataCite) identifiers on datasets &
GUIDs on records to measure impact• Technically sustainable & scalable• Default open licensing (CC-Zero, CC-BY, CC-BY-NC)
http://data.nhm.ac.uk
CKAN – the technical foundation for the portal• Enterprise, open source data portal platform• Developed by Open Knowledge Foundation• Used by 31 national governments, 74
regional authorities, academia & large commercial organisations
• Key featureso Publish & find datasetso Store & manage large datao Robust APIo Customise & extendo Sustainable
http://ckan.org/ e.g. http://data.gov.uk/
Primary views of each NHM dataset
Point map Grid map Heat map
Statistical overviewFilterable table
Dataset & data record citation• DataCite DOIs on every dataset• Stable URI (UUID) on every record• Prior identifiers aliased &
disambiguated• Citation encouraged with clear
statements at dataset & record level• Allows us to track cited usage• Dynamic DOI’s on subsets coming soon
Dataset DOI Specimen URI
Traffic-light data quality indicators (via GBIF)
Via GBIF API
Major errors
Minor errors
No errors
nb. similar services offered by CRIA for Brazilian data
Potential errors highlighted & “corrected”
Assembly Video
doi: 10.3897/zookeys.481.8788
Step-by-step instructions
Supports deposition of other research datasets
Easy addition of new datasets (rapid & semi-automated)
1. Name the dataset
2. Upload / link the data file
3. Describe the data file
4. Theme & tag
5. Add additional resources
6. Temporal coverage
7. Geographic coverage
8. Save & finish
Data access & feedback
Extensive API
R integration
Link to data curator team
DwCA Downloads RDF (Linked Open Data)
Serving external data aggregators
GBIF iDigBio EOL
Vertnet CRIA
Data visualisations driven by API
DEMO DEMO DEMO
500,000,000(since Feb. 2015, excluding major aggregators)
Records downloaded
Data access & feedback
Extensive API
R integration
Link to data curator team
DwCA Downloads RDF (& Linked Open Data)
Tim Berners-Lee, the inventor of the Web and Linked Data initiator, suggested a 5-star deployment scheme for Open Data…
Availa
ble Structu
red Non-proprietary
URI’sLin
ked (L
OD)
What does a 5-star Data Portal mean?
LOD gives us the means to connect our data (i.e. graph queries across distributed datasets)
Top 200 collections holding institutions contributing specimen record to GBIF
Example 1: “what data are we publishing”
• What proportion of our collections are accessible / digitised?
• What biases exiting in our digitised collections?
• How much taxonomic redundancy exists in our collections?
Useful for policy setting:- Planning digitisation strategies
(why should we all be digitising the same taxa first)- Identifying institutional collections strengths
(outside our community these are often not known)- What is ‘unique’ in our collections
(taxonomically, geospatially, temporally)- Disaster planning
(how many institutions hold the same material)
What collections are held globally?Where are these specimens from?
There are huge gaps and biases in what & where about our collections & where these collections are from
Top 200 collections(scaled by size)
Specimen country origin(darker is more )
Our results are very incomplete,constrained by what we’ve digitised
Size of collection
Proportion digitised
RBGE
RBGK
NHMMNHN
RMCA
RBINS
Very small proportions of our collections are digitally accessibleWe don’t publish the overall size of our collections in a machine readable way
Example 2: exploring ecological interactions
• Specimen data is one dimension of our collections
• We need to know how organisms interactE.g. Predator-prey, pollinator-pollenated, host-parasite
• Museums have lots of this data
NHM Interactions data:• Louse-host (12,000+)• Helminth host-parasite (250,000+)• Also large datasets: Coleoptera feeding on
dipterocarp seeds, butterfly host-plants, British mammal-flea associations, bee flower pollinators, several parasitic wasp datasets, ….
Increasingly published as RDF via NHM Data Portal
Global Biotic Interactions (GloBI) Database
• By Jorrit Poelen & colleagues• Collates interaction datasets• Currently >1.9M interactions• EOL pulls these into Species Pages• NHM Portal creates a combined
dataset to feed GloBI• Produces Linked Open Data
– Create beautiful visualisations
http://www.globalbioticinteractions.org/
• Predatory interactions for Eurythenes gryllus
• Visualisations highlight number, frequency & type of interaction
GloBI’s Interaction Browser
https://blog.globalbioticinteractions.org/2014/03/21/exploring-antarctic-
interactions-using-globis-interaction-browser/
Create beautiful visualisations with custom R scripts and existing libraries
(e.g., igraph, Reol, rgdal)https://blog.globalbioticinteractions.org/
2014/06/06/a-food-web-map-of-the-world/
Conclusions
• Data portals like the NHM Portal allow us to contribute and reflect our data through the lens of specialist aggregators
• GBIF & GloBI are specialist aggregators serving LOD• LOD allows us to combine big datasets to address new questions
– Tracking interactions & distribution of disease vectors– Predicting crop pests, via the distribution and interactions of pests of crop wild relatives
Next Steps• Continue Portal development & encourage institutional adoption• Consolidate NHM ecological interaction datasets• Publish combined dataset on the NHM Data Portal• GloBI to harvest the dataset and publish linked open data• Develop visualisations for key NHM datasets
Acknowledgements
Ben Scott – Portal Engineer & Architect
Ed Baker – Data Researcher
Laurence Livermore - Project Management
Matt Woodburn – Data Architect
Vince Smith – SRO / Coordinator
Top Related