1. A Connected Digital Biomedical Research Enterprise with Big Data Belinda Seto, Ph.D. Deputy Director National Eye Institute

2. What is it?  Digital research assets: data, workflow, publications, software  To connect these assets  Unique identifiers or tags  Annotation  Community-developed standards  Interfaces

3. Benefits  Increase scientific productivity  Enhance collaborations  Foster creativity: new tools, algorithms, methods, modeling  Enable new discoveries  Improve interoperability  Facilitate reproducibility

4. Gene Expression Data Barrett T et al. Nucl. Acids Res. 2013;41:D991-D995 Published by Oxford University Press 2012. Volume Velocity Variety

5. Gene Expression Omnibus  A public repository (NLM) of microarray, next generation sequencing and functional genomic data  Web-based interface and apps for query and data download

6. Myriad Data Types Other ‘Omic ImagingPhenotypic Clinical Genomic Exposure

7. Making Big Data Functional  Engender interdisciplinary approach to data collection and analysis by integrating scientific, algorithmic, and computational work  Drive functional data collection and analysis that has practical value in determining risk alleles

8. Integration of Data  Opportunities: Understanding biology across scales, from molecules to population  Challenges: need access to primary data and processed data, machine- readable metadata, tools to reduce dimensionality

9. Integration of Disparate Data Types: Brain Images with Genomic

10. Brain measures versus epidemiological studies to find genetic variants that directly affect the brain DIFFICULT EASIER ? May require 10,000-30,000 people e.g., the Psychiatric Genetics Consortium studies Gene variants (SNP’s) may affect brain measures directly, many brain measures relate to disease status.

11. Finding Genetic Variants Influencing Brain Structure … CTAGTCAGCGCT CTAGTAAGCGCT CTAGTCAGCGCT SNP C/C A/C A/A Intracranial Volume PhenotypeGenotype Association

12. Genome-Wide Association Studies (GWAS)  Identify loci for phenotypes or diseases using genotyping arrays throughout entire genome  Study association of polymorphisms with complex human traits  Meta-analysis across multiple studies

13. One SNP “Candidate gene” approach e.g., BDNF Screening 500,000 SNPs – 2,000,000 SNPs Position along genome NIH-funded database of genotypes and phenotypes enabling searches to find where in the genome a variant is associated with a trait. Genome-wide Association Study -log 10 (P-value) C/C A/C A/A Intracranial Volume

14. Applications of GWAS  Identify genetic variants that affect brain measures: volumetric, fiber integrity, connectivity  Risk genes  Early biomarkers of disease

15. What is a risk gene? - A common genetic variant related to a brain measure, or a disease, or a trait such as obesity, found by searching the genome 23 pairs of chromosomes In a particular part of the chromosome 5 there are many genes Within a gene there are exons, introns, and SNPs Single Nucleotide Polymorphism (SNP) 99.9% of DNA is the same for all people - DNA variation causes changes in predisposition to disease, and brain structure. One type of variation is a single nucleotide polymorphism (SNP)- Single letter change in the DNA code

16. GRIN2B Risk Allele  Glutamate receptor, signaling pathway  Genetic polymorphism of GRIN2B gene  Associated with reductions of brain white matter integrity  Bipolar disorder  Obsessive compulsive disorder

17. Jason L. Stein 1, Xue Hua PhD 1, Jonathan H. Morra PhD 1, Suh Lee 1, April J. Ho 1, Alex D. Leow MD PhD 1,2, Arthur W. Toga PhD 1, Jae Hoon Sul 3, Hyun Min Kang 4, Eleazar Eskin PhD 3,5, Andrew J. Saykin PsyD 6, Li Shen PhD 6, Tatiana Foroud PhD 7, Nathan Pankratz 7, Matthew J. Huentelman PhD 8, David W. Craig PhD 8, Jill D. Gerber 8, April Allen 8, Jason J. Corneveaux 8, Dietrich A. Stephan 8, Jennifer Webster 8, Bryan M. DeChairo PhD 9, Steven G. Potkin MD 10, Clifford R. Jack Jr MD 11, Michael W. Weiner MD 12,13, Paul M. Thompson PhD 1,*, and the ADNI (2010). Genome-Wide Analysis Reveals Novel Genes Influencing Temporal Lobe Structure with Relevance to Neurodegeneration in Alzheimer's Disease, NeuroImage 2010. GRIN2b genetic variant is associated with 2.8% temporal lobe volume deficit GRIN2b is over-represented in AD - could be considered an Alzheimer’s disease risk gene - needs replication

18. Jason L. Stein 1, Xue Hua PhD 1, Jonathan H. Morra PhD 1, Suh Lee 1, April J. Ho 1, Alex D. Leow MD PhD 1,2, Arthur W. Toga PhD 1, Jae Hoon Sul 3, Hyun Min Kang 4, Eleazar Eskin PhD 3,5, Andrew J. Saykin PsyD 6, Li Shen PhD 6, Tatiana Foroud PhD 7, Nathan Pankratz 7, Matthew J. Huentelman PhD 8, David W. Craig PhD 8, Jill D. Gerber 8, April Allen 8, Jason J. Corneveaux 8, Dietrich A. Stephan 8, Jennifer Webster 8, Bryan M. DeChairo PhD 9, Steven G. Potkin MD 10, Clifford R. Jack Jr MD 11, Michael W. Weiner MD 12,13, Paul M. Thompson PhD 1,*, and the ADNI (2010). Genome-Wide Analysis Reveals Novel Genes Influencing Temporal Lobe Structure with Relevance to Neurodegeneration in Alzheimer's Disease, NeuroImage, 2010. GRIN2b genetic variant associates with brain volume in these regions; 2.8% more temporal lobe atrophy

19. Alzheimer’s risk gene carriers (CLU-C) have lower fiber integrity even when young (N=398), 50 years before disease typically hits Voxels where CLU allele C (at rs11136000) is associated with lower FA after adjusting for age, sex, and kinship in 398 young adults (68 T/T; 220 C/T; 110 C/C). FDR critical p = 0.023. Left hem. on Right Braskie et al., Journal of Neuroscience, May 4 2011

20. Effect is even stronger for carriers of a schizophrenia risk gene variant, trkA-T (N=391 people) a. p values indicate where NTRK1 allele T carriers (at rs6336) have lower FA after adjusting for age, sex, and kinship in 391 young adults (31 T+; 360 T-). FDR critical p = 0.038. b. Voxels that replicate in 2 independent halves of the sample (FDR-corrected). Left is on Right. Braskie et al., Journal of Neuroscience, May 2012

21. Neural Fiber Integrity Fractional Anisotropy  Applied to diffusion tensor MRI  Eigen = 0 means diffusion is totally unrestricted  Eigen = 1 means diffusion is restricted to only one direction  FA measures fiber density, axonal diameter, or myelination of white matter

22. Kohannim O, et al. Predicting white matter integrity from multiple common genetic variants. Neuropsychopharmacology 2012, in press. COMT HFE CLU NTRK1 ErbB4 BDNF SNP’s can predict variance in brain integrity Neuro-chemical genes Neuro- developmental genes Neuro- degenerative risk genes A significant fraction of variability in white matter structure of the corpus callosum (measured with DTI) is predictable from SNPs;

23. Big Data  26,000 whole brain MR images  > 500,000 single nucleotide polymorphism (SNP)  Analyze each voxel of the entire brain and search for genetic variants of the whole genome at each brain voxel  Select only the most associated SNP at each voxel, by analyzing P-values through an inverse beta transformation

24. Genetic clustering boosts GWAS power 1.Many top hits now reach genome-wide significance (N=472) and replicate 2.Several SNPs affect multiple ROIs 3.Can form a network of SNPs that affect similar ROIs 4.It has a small-world, scale-free topology (for more, see Chiang et al., J. Neurosci., 2012)

25. Population level Data Integration: Electronic Medical Records, Genotypes and Phenotypes

26. eMERGE  Goal: research to combine DNA biorepositories with EMR for large- scale association studies of genetics and phenotypes; to incorporate genetic variants into EMG for use in clinical care

27. Network Members

28. eMERGE Innovation  Algorithms for electronic phenotyping of clinical conditions identified in EMR  Discoveries of genetic variants in biorepository samples

29. Big Data to Knowledge