Informatics Infrastructure for Clinical Genomics
- 1. © 2013 New York Genome Center 1PRIVILEGED & CONFIDENTIAL Challenges of Building Informatics Infrastructure for Clinical Genomics TOBY BLOOM, PH.D. OCTOBER 5, 2013
- 2. © 2013 New York Genome Center 2PRIVILEGED & CONFIDENTIAL COLLABORATION OF 12 FOUNDING INSTITUTIONS
- 3. © 2013 New York Genome Center 3PRIVILEGED & CONFIDENTIAL ASSOCIATE MEMBERS More To Come Objective: Extend NYGC value by expanding collaborative network and enriching scientific community
- 4. © 2013 New York Genome Center 4PRIVILEGED & CONFIDENTIAL NEW YORK GENOME CENTER Fostering collaborative genomics research across New York area Central sequencing, informatics infrastructure, analysis services for member institutions & others Collaborative research projects Focus on clinical genomics Still in “start-up” mode Bob Darnell – President & Scientific Director HHMI Prof. of Neuroscience at Rockefeller University First faculty member – Tuuli Lapalainen Joint w/ Columbia
- 5. © 2013 New York Genome Center 5PRIVILEGED & CONFIDENTIAL FROM THE INFORMATICS PERSPECTIVE Build an informatics environment that fosters large collaborative projects and provides the infrastructure for large-scale clinical genomics studies Compute and storage resources Informatics tools and methods Privacy enforcement Computational biology services Standard large datasets Reference databases Submission and retrieval services
- 6. © 2013 New York Genome Center 6PRIVILEGED & CONFIDENTIAL CLINICAL INFORMATICS INFRASTRUCTURE- CURRENT CHALLENGES Clinical data management Analysis integrating disparate, longitudinal data sources Pipeline performance turn-around vs throughput Pipeline quality and validation SECURITY and PRIVACY Tracking IRBs, informed consents, DAC approvals HIPAA, HiTech compliance 21 CFR Part 11 CLIA
- 7. © 2013 New York Genome Center 7PRIVILEGED & CONFIDENTIAL STARTING “BACKWARDS”: CLINICAL DATA Working with 6 hospitals on a federated, longitudinal clinical data warehouse 2.5 million patients to start Could grow to 8 million Basic, anonymized clinical data stored at NYGC Connected to hospital data warehouses for more complex queries
- 8. © 2013 New York Genome Center 8PRIVILEGED & CONFIDENTIAL DATA STORED AT NYGC At least 5 years of longitudinal records for each Fully anonymized, HIPAA safe-haven data Available for cohort identification and retrospective studies Patient-matching across institutions Tagged with cohort consents Tagged by availability of genomic data
- 9. © 2013 New York Genome Center 9PRIVILEGED & CONFIDENTIAL PROSPECTIVE FUNCTIONALITY Study-specific data marts With informed consent Connections back to hospital warehouse edge servers for more complex data Collection of study-specific clinical research data Connections to genomic data
- 10. © 2013 New York Genome Center 10PRIVILEGED & CONFIDENTIAL THE CLINICAL REPOSITORY
- 11. © 2013 New York Genome Center 11PRIVILEGED & CONFIDENTIAL DATA TYPE INTEGRATION – COMPLEX AND GROWING ResultsClinical Data (Longitudinal) Genome RNA-Seq (Longitudinal) Microbiome ChIP-Seq Personal- Reported Data(FitBit) Reference Filters
- 12. © 2013 New York Genome Center 12PRIVILEGED & CONFIDENTIAL DATA INTEGRATION: ONE EXAMPLE Auto-immune disease project 4 hospitals and NYGC Cross-disease At least weekly blood collection More frequent around flares At least monthly clinical evaluations More frequent around flares Continuous personal device data Correlate changes in expression w/ changes in clinical metrics, and changes in personal device data streams
- 13. © 2013 New York Genome Center 13PRIVILEGED & CONFIDENTIAL DATA INTEGRATION: THE COMPLEXITY CONTINUES TO GROW Genomic WGS WES RNA-Seq ChIP-Seq Clip-Seq RRBS Longitudinal? EHR* Different dictionaries Different standards Doctors notes Personal device Patient-reported Standard datasets, references, ….
- 14. © 2013 New York Genome Center 14PRIVILEGED & CONFIDENTIAL LONGITUDINAL MULTI-MODAL DATA Different types have different event frequencies Do baselines exist? How many variables?
- 15. © 2013 New York Genome Center 15PRIVILEGED & CONFIDENTIAL ACCURATE ANALYSIS METHODS Different tools produce different results That’s one thing in TCGA or 1000G It’s another for clinical pipelines How many tools do we need to run in parallel? How do we know whether what we find is the cause? How sure are we that the associated treatment is right? If that variant has other implications, what do we do? Yes, we can validate But how many rounds? At what cost?
- 16. © 2013 New York Genome Center 16PRIVILEGED & CONFIDENTIAL ANOTHER EXAMPLE Rare pediatric cancer Research study Tumor-normal- recurrence Some of the kids are still alive RNA-Seq and WGS 3 fusion tools Before we found the fusion 3 structural variant tools to confirm the deletion No one tool found all of them 2 split read tools to find the exact breakpoints Nico Robine & Anne-Katrin Emde
- 17. © 2013 New York Genome Center 17PRIVILEGED & CONFIDENTIAL METHOD ACCESS RIGHTS Academic only? Non-profit only? Commercially available Possibly different licenses or different versions So multiple different pipelines for different users That get different results!!
- 18. © 2013 New York Genome Center 18PRIVILEGED & CONFIDENTIAL HIGH-PERFORMANCE: WHAT’S THE GOAL? A single patient in a hospital bed is not the same as 1000 genomes in a research study. How do we ensure that pipelines run fast enough? For large research studies, Want high-throughput - fastest way to get all the samples out For clinical samples. Want fast turn-around Can’t get both
- 19. © 2013 New York Genome Center 19PRIVILEGED & CONFIDENTIAL DATA SECURITY AND PRIVACY Compliance HIPAA HiTech 21 CFR Part 11 FISMA IRB approvals Informed Consents Data Access Committees Who decides? Who enforces?
- 20. © 2013 New York Genome Center 20PRIVILEGED & CONFIDENTIAL HOSPITAL REGULATIONS No cloud No public transports – like globus online All data encrypted Can you really run a wgs pipeline with the reads encrypted? BTW – hardware-assisted decryption takes 3 hours/BAM
- 21. © 2013 New York Genome Center 21PRIVILEGED & CONFIDENTIAL PI’S DON’T UNDERSTAND INFORMED CONSENT Who they can give their data to What data they can use, if they can find it on the file system So the system has to monitor Recent example: We were asked to host a large dataset First told it was publicly available data Then corrected to “broadly consented” We asked to see the IRB and an informed consent THEN they realized they needed IRB approval for us to host the data, and the IRB wanted to see our security plans.
- 22. © 2013 New York Genome Center 22PRIVILEGED & CONFIDENTIAL VERY GRANULAR ACCESS CONTROL Access control is not per patient Multiple studies per patient Different Pis on each Not per project Informed consents still have check boxes So some researchers can get access to some samples but not others from a project So need the ability to have same file in multiple projects
- 23. © 2013 New York Genome Center 23PRIVILEGED & CONFIDENTIAL PROVIDING A SECURE ENVIRONMENT All data sits on shared storage But that storage is isolated All access goes through an invisible “security gate” that allows researchers to see only the data they have rights to: Their data Data for which they have data use authorization Reference data Public data Catalog of all data Needs to be fast and transparent They need to work on virtual machines, so no data on local disk or in memory is accidently shared.
- 24. © 2013 New York Genome Center 24PRIVILEGED & CONFIDENTIAL MAINTAINING DATA PRIVACY StorageCompute Access Control
- 25. © 2013 New York Genome Center 25PRIVILEGED & CONFIDENTIAL SUMMARY Integrating with clinical data is not easy Infrastructure is complex Need to understand analysis methods better Need to understand security better
- 26. © 2013 New York Genome Center 26PRIVILEGED & CONFIDENTIAL 101 AVENUE OF THE AMERICAS COLLABORATIVE HUB We’re hiring J Computational Biologists Software Engineers Project Mgrs Lab Personnel Principal Investigators Faculty Joint Appts ………..
- 27. © 2013 New York Genome Center 27PRIVILEGED & CONFIDENTIAL ACKNOWLEDGEMENTS NYGC Bob Darnell Vlada Vacic Nicolas Robine Avinash Abhiyankar Uday Evani Anne-Katrin Emde James Spencer Nina Lapchyk Soren Germer Dayna Oschwald Yanlei Diao (UMass) Deborah Estrin (Cornell Tech) Rainu Kaushal (Weil-Cornell) George Hripsak(Columbia) Tom Check (Healthix) Tom Campion (Cornell) Parsa Mirhaji ( Einstein) Dana Orange (Rockefeller) Nathaniel Novod (Broad) Seva Kashin (Mass General) Leslie Greengard (Simons Center) Alex Lash (Simons Center)