Correlation globes of the exposome 2016
1 like375 views
This document discusses developing exposome correlation globes to map associations between exposures and phenotypes. It summarizes work analyzing replicated correlations between over 250 quantitative exposures measured in NHANES participants to create a globe visualization. The analysis found that while the exposome correlations were dense, with around 3% of pair-wise correlations replicated between cohorts, the correlations were modest in absolute size. The exposome globes could help contextualize exposome-wide association studies and identify co-occurring exposures.
Correlation globes of the exposome 2016
- 1. Development of exposome correlation globes to map out exposure-phenotype associations Chirag J Patel (and Arjun K Manrai) International Society of Exposure Science Utrecht 2016 10/10/16 [email protected] @chiragjp www.chiragjpgroup.org
- 2. Pesticides Pollutants Vitamins Nutrients Infectious Agents Diabetes Body Mass Index Time-to-Death Gene expression Telomere length PhenomeExposome How is the exposome associated with the phenome?
- 3. How is the exposome associated with the phenome?: Searches for exposures in telomere-length IJE, 2016 0 1 2 3 4 −0.2 −0.1 0.0 0.1 0.2 effect size −log10(pvalue) PCBs FDR<5% Trunk Fat Alk. PhosCRP Cadmium Cadmium (urine)cigs per day retinyl stearate VO2 Maxpulse rate shorter telomeres longer telomeres adjusted by age, age2, race, poverty, education, occupation median N=3000; N range: 300-7000 Co-exposure plays a role in signal and association
- 4. Number of potential correlates complicates the association between exposure and phenome IJE 2012 Sci Trans Med 2011 Pesticides Pollutants Vitamins Nutrients Infectious Agents Diabetes Body Mass Index Time-to-Death Gene expression Telomere length PhenomeExposome
- 5. Arjun Manrai “This is what I call ‘indecent exposome'”
- 6. Does Bradford-Hill apply?: Sheer number of correlations of the exposome have implications for causal research Stat Med, 2015
- 7. Does Bradford-Hill apply?: Sheer number of correlations of the exposome have implications for causal research, for example: (1) Strength of associations: correlation & p-values (2) Consistency: observed in different situations? (3) Specificity: do one-to-one associations exist?
- 8. Estimating correlations in E: What does this buy us in conducting EWAS-like investigations? (1) Effective number of variables to test in EWAS Criterion 1: Significance (p-values) (2) Mapping/documenting EWAS associations Criterion 3: Specificity (3) Assessing correlations due to model choice Criterion 2: Consistency
- 9. Estimating exposome ρ: NHANES participants have >250 quantitative exposures assayed in serum and urine and >500 via self-report! Nutrients and Vitamins e.g., vitamin D, carotenes Pesticides and pollutants e.g., atrazine; cadmium; hydrocarbons Infectious Agents e.g., hepatitis, HIV, Staph. aureus Plastics and consumables e.g., phthalates, bisphenol A Physical Activity e.g., steps
- 10. Estimating exposome ρ: Replicated rank correlations between exposures and visualized with a globe ’99-’00 ’01-’02 ’03-’04 ’05-’06 289 357 456 313 | E | 575 35,835 56,557 80,401 47,203 81,937 | ρ(e1,e2) |cohorts N:10-10K FDR(e1,e2) < 5% in >1 cohorts? (Benjamini-Hochberg) Permutation-based p-values Replicated(e1,e2) are linked e1 e2e4 e3 http://circos.ca ρ>0ρ<0
- 11. Estimating exposome ρ: E correlations are concordant between independent cohorts ‘99-’00 ‘01-’02 ‘03-’04 ‘05-’06 ‘99-’00 1 0.84 0.84 0.92 ‘01-’02 1 0.82 0.93 ‘03-’04 1 0.94 ‘05-’06 1 2,656 out of 81,937 (3%) pair-wise correlations (FDR < 5% in > 1 cohort) N:10-10K
- 12. The E correlation globe is dense (2,700 out of 81K), but correlations are modest in absolute value (median: 0.45). 0.00 0.25 0.50 0.75 1.00 0.0 0.4 0.8 |Correlation| Cumulativefraction all correlations q<=0.05 >1 surveys q<=0.05 >2 surveys (replicated) FDR<5% FDR<5% in >1 cohort
- 13. Replicated E correlations are modest in size and are mostly positive 0 5 10 15 -1.0 -0.5 0.0 0.5 1.0 Correlation Percent ρ>0ρ<0
- 14. Estimating correlations in E: What does this buy us in conducting EWAS-like investigations? (1) Effective number of variables to test in EWAS (2) Mapping/documenting EWAS associations (3) Assessing correlations due to model choice Criterion 1: Significance (p-values) Criterion 3: Specificity Criterion 2: Consistency
- 15. Estimating correlations in E: Effective number of variables in your data - You measure M: are they all independent? Meff ≤ M Meff : 1 + (M - 1) (1 - Variance(L)/M) L: eigenvalues JECH, 2014 M: number of variables Meff: effective number co-exposure correlation
- 16. Meff influences signal to noise and power!
- 17. Dense ρ influences the number of effective variables (Meff) in NHANES JECH, 2014 National Health and Nutrition Examination Survey (NHANES)
- 18. Estimating correlations in E: What does this buy us in conducting EWAS-like investigations? (1) Effective number of variables to test in EWAS (2) Mapping/documenting EWAS associations (3) Assessing correlations due to model choice Criterion 1: Significance (p-values) Criterion 3: Specificity Criterion 2: Consistency
- 19. Estimating exposome ρ: Replicated rank correlations between exposures and visualized with a globe ’99-’00 ’01-’02 ’03-’04 ’05-’06 289 357 456 313 | E | 575 35,835 56,557 80,401 47,203 81,937 | ρ(e1,e2) |cohorts N:10-10K FDR(e1,e2) < 5% in >1 cohorts? (Benjamini-Hochberg) Permutation-based p-values Replicated(e1,e2) are linked e1 e2e4 e3 http://circos.ca ρ>0ρ<0
- 20. Visualizing replicated E correlations with an exposome globe Arranging exposures by category 198 14 36 82 3 17 47 59 25 31 51 12 7 38 65 7 10 8 15 12 7 22017 6
- 21. Visualizing replicated E correlations with an exposome globe exposures linked to cotinine, a metabolite of nicotine ρ>0: red ρ<0: blue
- 22. Visualizing replicated E correlations with an exposome globe 2,656 (out of 81,937) pair-wise correlations ρ>0: red ρ<0: blue
- 23. Telomere Length All-cause mortality http://bit.ly/globebrowse Interdependencies of the exposome: Telomeres vs. all-cause mortality
- 24. Browse these and 82 other phenotype-exposome globes! http://www.chiragjpgroup.org/exposome_correlation https://github.com/chiragjp/exposome_correlation
- 25. Estimating correlations in E: What does this buy us in conducting EWAS-like investigations? (1) Effective number of variables to test in EWAS (2) Mapping/documenting EWAS associations (3) Assessing correlations due to model choice Criterion 1: Significance (p-values) Criterion 3: Specificity Criterion 2: Consistency
- 26. On dense ρ and exposome globes: Discussion and Future Directions •E ρ are dense (~3% of links replicated!) but modest in correlation size •Visualize and identify co-occuring E •Contextualize EWAS findings Demographic Food(Recall) PhysicalActivity Nutrients Smoking Drugs Furans Dioxins PCBs Pesticides Diakyls PFCs Phenols Phthalates Bacteria Virus Urinary_Dim ethylarsonic_acid number_of_days_since_quit Beta-hexachlorocyclohexane trans-b-carotene 3,3,4,4,5,5-hxcb M ercury,_urine Oxychlordane um ,_urine 1,2,3,4,6,7,8,9-ocdd ercury,_inorganic 1,2,3,4,6,7,8-hpcdd Hexachlorobenzene g-tocopherol Retinyl_palm itate PCB138_&_158 PCB196_&_203 Vitam in_D ny,_urine Vitam in_C 1,2,3,4,7,8-hxcdd 1,2,3,6,7,8-hxcdd 1,2,3,7,8,9-hxcdd Cadm ium ,_urine Retinyl_stearate 1,2,3,4,7,8-hxcdf 1,2,3,6,7,8-hxcdf Trans-nonachlor Heptachlor Thallium ,_urine b-cryptoxanthin ury,_total 1,2,3,7,8-pncdd Levofloxacin_1 2,3,4,7,8-pncdf Folate,_serum 3,3,4,4,5-pncb ine a-Tocopherol CD8_counts PCB170 Lead,_urine 2,3,7,8-tcdd Oxacillin_1 a-Carotene Cadm ium p,p-DDT p,p-DDE PCB105 PCB118 PCB156PCB157 PCB167 PCB146PCB153PCB172 PCB177 PCB178 PCB180 PCB183PCB187 PCB194 PCB199 Dieldrin PCB66PCB74 PCB99 Retinol white black Lead Age •Estimate and report Meff, number of independent variables •Estimate and report VoE, how correlations change due to model choice
- 27. On dense ρ and exposome globes: Discussion and Future Directions •E ρ are dense (~3% of links replicated!) but modest in correlation size •Identify confounding variables? •Ascertain E globes with respect to time in different populations! Demographic Food(Recall) PhysicalActivity Nutrients Smoking Drugs Furans Dioxins PCBs Pesticides Diakyls PFCs Phenols Phthalates Bacteria Virus Urinary_Dim ethylarsonic_acid number_of_days_since_quit Beta-hexachlorocyclohexane trans-b-carotene 3,3,4,4,5,5-hxcb M ercury,_urine Oxychlordane um ,_urine 1,2,3,4,6,7,8,9-ocdd ercury,_inorganic 1,2,3,4,6,7,8-hpcdd Hexachlorobenzene g-tocopherol Retinyl_palm itate PCB138_&_158 PCB196_&_203 Vitam in_D ny,_urine Vitam in_C 1,2,3,4,7,8-hxcdd 1,2,3,6,7,8-hxcdd 1,2,3,7,8,9-hxcdd Cadm ium ,_urine Retinyl_stearate 1,2,3,4,7,8-hxcdf 1,2,3,6,7,8-hxcdf Trans-nonachlor Heptachlor Thallium ,_urine b-cryptoxanthin ury,_total 1,2,3,7,8-pncdd Levofloxacin_1 2,3,4,7,8-pncdf Folate,_serum 3,3,4,4,5-pncb ine a-Tocopherol CD8_counts PCB170 Lead,_urine 2,3,7,8-tcdd Oxacillin_1 a-Carotene Cadm ium p,p-DDT p,p-DDE PCB105 PCB118 PCB156PCB157 PCB167 PCB146PCB153PCB172 PCB177 PCB178 PCB180 PCB183PCB187 PCB194 PCB199 Dieldrin PCB66PCB74 PCB99 Retinol white black Lead Age •What are the essential nodes of the network?
- 28. On dense ρ and exposome globes: For papers, see: https://paperpile.com/shared/0SnSa9 Ioannidis, John P. A. 2016. Statistics in Medicine 35 (11): 1749–62. Patel, Chirag J., et al 2013. International Journal of Epidemiology 42 (6). IEA: 1795–1810. Patel, Chirag J., et al 2012. International Journal of Epidemiology 41 (3): 828–43. Patel, Chirag J., and Arjun K. Manrai. 2015. Pacific Symposium on Biocomputing., 231–42. Ioannidis, John P. A. 2009. Science Translational Medicine 1 (7). Patel, Chirag J., et al. 2015. Journal of Clinical Epidemiology 68. Patel, Chirag J., and John P. A. Ioannidis. 2014. JAMA: 311 (21). 2173–74. Smith, G. D., et al. 2007. “PLoS Medicine 4: e352. Tu-SY-A4: The Exposome: From concept to practice - IV
- 29. Harvard DBMI Isaac Kohane Susanne Churchill Stan Shaw Jenn Grandfield Sunny Alvear Michal Preminger Harvard Chan Hugues Aschard Francesca Dominici Chirag J Patel [email protected] @chiragjp www.chiragjpgroup.org NIH Common Fund Big Data to Knowledge Acknowledgements Stanford John Ioannidis Atul Butte (UCSF) RagGroup Chirag Lakhani Adam Brown Danielle Rasooly Arjun Manrai Erik Corona Nam Pho Jake Chung ISES Co-exposures Tom Webster Arjun Manrai