Open Data in Biology

Tim Hubbard, @timjph

First International Open Economics Workshop

17th December 2012

Biology is a grand project

• Build complete models of biological systems
• Intelligent drug development
• disease prediction
• modelling individuals (e.g. virtual human, ITFoM)

​

• Feels like should be parallel with economics, modelling economic systems

​

​

​

​

​

Data is organised towards this goal

• Centralised repositories for raw data
• one data type, one repository
• mandatory submission linked to publication
• Infrastructure to organise raw data for access
• human genome presented to user as whole chromosomes instead of thousands of fragments
• Curated databases of biological objects
• supported by evidence from raw data repositories

​

• First repositories >40 years old
• 1,000s of full time staff supporting infrastructure distributed worldwide
• Sanger/EBI alone >30 peta bytes of data

​

​

​

​

​

Data, Databases & Bioinformatics

​

​

​

​

Researchers

​

Data

Resources

Repositories

Global Infrastructure

Curation

Website

Data mining

APIs

Downloads

​

​

​

​

Researchers

​

Submission

​

Policies

Systems

​

​

Sustainable

funding

Reuse

​

Discoverability

Easy of use (+access)

Scientific Record

Complete sets of components

Human genome race��won by public project��open access for all

Scale

• Human genome = 6 billion letters
• encodes ~30,000 genes, 4 million switches
• Human individual = 100 trillion cells
• more than 200 cell types
• Human population = 7 billion

​

Scale of data

• Size of current raw DNA database
• 400 trillion letters of DNA
• 3.4 trillion DNA items

​

• Project size
• UK Biobank = 500,000 individuals
• Plan for NHS = sequence 100,000 in 3-5 years

​

Data infrastructure for biology

• Enforces standards for submission
• Grants contain ‘data sharing’ section
• Enables rapid exploration of new questions
• Assumption that if you have a new idea at 2am you can immediately access the data to test it
• Structured for reproducibility
• Unique identifiers
• Versioned data items and datasets
• Widespread use of ontologies
• No licenses restricting use of data allowed

Sanger implementation of �Data Sharing

• Previously
• Genome sequence (human consented + model organisms)
• Today
• Genome sequence, genotypes (some disease related), phenotype data (models), high throughput assay data (transcriptomics), WT Open Access (OA) publishing policy

​

• Data sharing committee
• Data sharing policy
• Tracking of compliance

Data sharing issues for institutes

• Tracking compliance needs to be proactive
• Pragmatic assessment of value of intermediate data

​

• WTSI approach to pre-publication data release
• Raw data automatically, immediately deposited to repositories
• Intermediate analysis provided via institute websites
• Final analysis outputs (e.g. linked with publication) submitted to appropriate database, repository

Ensuring data submission

• Include potential IT costs in grant application
• Submit metadata to repository
• Obtain Accession number from repository
• Collect Data
• Submit Data to repository

New policy: sanctions for non-compliance

• Current OA compliance levels “unacceptable”
• New sanctions:
1. In End of Grant Report all papers listed must be OA. If not the final payment on the grant (typically 10%) will be withheld
2. Non-compliant publications will be discounted as part of a researcher’s track record in any renewal of an existing grant or new grant application
3. Trust-funded researchers will need to ensure that all publications associated with their Wellcome-funded research are OA before any funding renewals or new grant awards will be activated
• Sanctions aimed at changing behaviour

Data Sharing

• Open to all
• Human Genome Projects where subject consented: Hapmap, 1000 genomes
• Repository: Genbank, ENA, DDBJ (INSDC)
• Managed access (must be bona fide researcher)
• Genetic data for disease cohorts, with phenotypes
• Repository: DbGaP, EGA (Encrypted distributions etc.)
• Currently very limited access
• Patient records
• Repository: Clinical Practice Research Datalink (CPRD)(formally RCP) (anonymised data)

Traditional: (Honest Broker)

Data set A

Researcher

“Run X on A & B”

​

​

​

​

​

​

​

​

Data set B

Results

​

​

​

​

“Request A & B data set”

Algorithm X

Data set combination

and anonymisation

process

Honest Broker Model:

• Identifiable data sets held securely by Honest Broker
• Anonymised data sets generated on demand and distributed to researchers
• Researcher applies custom algorithm or analysis to generate results

Honest Broker

Anonymised data set

Proposed: (SVM)

Data set A

Researcher

“Run X on A & B”

​

​

​

​

​

​

​

​

Data set B

Results

“Run VM on A & B”

Summary data only

via output API

(no raw data)

Algorithm X

Secure Virtual machine (SVM):

• Access to raw data by algorithm can only be via APIs contained in template VM
• Export of summary data can only be via output APIs contained in template VM

Honest Broker (with local cloud)

Download VM Template

Changing sources of biological data

Healthcare Professional

Component 4

Individual query analysis

Component 3

Additional clinical annotation

Component 2

Genotype and Phenotype relationship capture

Component 1

Human sequence data repositories

Component 5

Electronic Health Record &

Personal Genome Sequence

​

Data from Patients

(NHS)

​

Data from Collections

(Research Institutes)

E-Health, Genomic Medicine & Linked Data

Healthcare Professional

Component 4

Individual query analysis

Component 3

Additional clinical annotation

Component 2

Genotype and Phenotype relationship capture

Component 1

Human sequence data repositories

Component 5

Electronic Health Record &

Personal Genome Sequence

​

Data from Patients (NHS)

​

CPRD

​

Data from Collections

(Research Institutes)

DH/NHS

DWP

HMRC

ONS

DfE, BIS

DJ, IC

​

ESRC

WT

MRC

Openness Privacy

• Can you have both?

​

Acknowledgements

Discussions with many at Sanger Institute, EBI, Wellcome Trust, NCBI, NHGRI, europePMC, Human Genome Strategy Group, Administrative Data Taskforce