Genetics Studies of Multivariate Traits

Size: px

Start display at page:

Download "Genetics Studies of Multivariate Traits"

Harold Mason
5 years ago
Views:

1 Genetics Studies of Multivariate Traits Heping Zhang Department of Epidemiology and Public Health Yale University School of Medicine Presented at Southern Regional Council on Statistics Summer Research Conference June 7, 2011 Heping Zhang (C 2 S 2, Yale University) SRCOS 1 / 61

2 Outline 1 Background Comorbidity Disorders, Genes and Covariates 2 Weighted Association Test Generalized Kendall s Tau Asymptotic Distribution and Power Application to WTCCC Bipolar Disorder Data 3 Maximum Weighted Association Test Asymptotic Distribution Simulations Application to COGA Data 4 Conclusions Heping Zhang (C 2 S 2, Yale University) SRCOS 2 / 61

3 Outline 1 Background Comorbidity Disorders, Genes and Covariates 2 Weighted Association Test Generalized Kendall s Tau Asymptotic Distribution and Power Application to WTCCC Bipolar Disorder Data 3 Maximum Weighted Association Test Asymptotic Distribution Simulations Application to COGA Data 4 Conclusions Heping Zhang (C 2 S 2, Yale University) SRCOS 3 / 61

4 Comorbidity Multiple disorders or illnesses occur in the same person, simultaneously or sequentially Source: Heping Zhang (C 2 S 2, Yale University) SRCOS 4 / 61

5 Comorbidity Source: aasets.lifehack.org Heping Zhang (C 2 S 2, Yale University) SRCOS 5 / 61

Comorbidity Is there a relationship between childhood ADHD and later drug abuse? See page 2.

It is also a topic about which we have insufficient information, so it remains a research priority for NIDA. This Research Report provides information on the state of the science in this area.

* To help explain this comorbidity, we need to first recognize that drug addiction is a mental illness.

6 Comorbidity Is there a relationship between childhood ADHD and later drug abuse? See page 2. from the director: Comorbidity is a topic that our stakeholders patients, family members, health care professionals, and others frequently ask about. It is also a topic about which we have insufficient information, so it remains a research priority for NIDA. This Research Report provides information on the state of the science in this area. Although a variety of diseases commonly co-occur with drug abuse and addiction (e.g., HIV, hepatitis C, cancer, cardiovascular disease), this report focuses only on the comorbidity of drug use disorders and other mental illnesses.* To help explain this comorbidity, we need to first recognize that drug addiction is a mental illness. It is a complex brain disease characterized by compulsive, at times uncontrollable drug craving, seeking, and use despite devastating consequences behaviors that stem from drug-induced changes in brain structure and function. These changes occur in some of the same brain areas that are disrupted in other mental disorders, such as depression, anxiety, or schizophrenia. It is therefore not surprising that population surveys show a high rate of co-occurrence, or comorbidity, between drug addiction and other mental illnesses. While we cannot always prove a connection or causality, we do know that certain mental disorders are established risk factors for subsequent drug abuse and vice versa. It is often difficult to disentangle the overlapping symptoms of drug addiction and other mental illnesses, making diagnosis and treatment complex. Correct diagnosis is critical to ensuring appropriate and effective treatment. Ignorance of or failure to treat a comorbid disorder can jeopardize a patient s chance of recovery. We hope that our enhanced understanding of the common genetic, environmental, and neural bases of these disorders and the dissemination of this information will lead to improved treatments for comorbidity and will diminish the social stigma that makes patients reluctant to seek the treatment they need. Nora D. Volkow, M.D. Director National Institute on Drug Abuse Comorbidity: Addiction and Other Mental Illnesses What Is Comorbidity? W hen two disorders or illnesses occur in the same person, simultaneously or sequentially, they are described as comorbid. Comorbidity also implies interactions between the illnesses that affect the course and prognosis of both. continued inside *Since the focus of this report is on comorbid drug use disorders and other mental illnesses, the terms mental illness and mental disorders will refer here to disorders other than substance use disorders, such as depression, schizophrenia, anxiety, and mania. The terms dual diagnosis, mentally ill chemical abuser, and co-occurrence are also used to refer to drug use disorders that are comorbid with other mental illnesses. Dr. Volkow, Director, NIDA: Comorbidity is a topic that our stakeholders patients, family members, health care professionals, and others frequently ask about. It is also a topic about which we have insufficient information, so it remains a research priority for NIDA. Source: Heping Zhang (C 2 S 2, Yale University) SRCOS 6 / 61

7 Possible Mechanisms for Comorbidity Mental disorder drug use disorder Drug use disorder { mental disorder mental disorder Common etiology drug use disorder Source: Heping Zhang (C 2 S 2, Yale University) SRCOS 7 / 61

8 Possible Mechanisms for Comorbidity Mental disorder drug use disorder Drug use disorder { mental disorder mental disorder Common etiology drug use disorder Source: Heping Zhang (C 2 S 2, Yale University) SRCOS 7 / 61

9 Possible Mechanisms for Comorbidity Mental disorder drug use disorder Drug use disorder { mental disorder mental disorder Common etiology drug use disorder Source: Heping Zhang (C 2 S 2, Yale University) SRCOS 7 / 61

10 Outline 1 Background Comorbidity Disorders, Genes and Covariates 2 Weighted Association Test Generalized Kendall s Tau Asymptotic Distribution and Power Application to WTCCC Bipolar Disorder Data 3 Maximum Weighted Association Test Asymptotic Distribution Simulations Application to COGA Data 4 Conclusions Heping Zhang (C 2 S 2, Yale University) SRCOS 8 / 61

11 Genotypes and Covariates Source: en.wikipedia.org; 2010.census.gov Heping Zhang (C 2 S 2, Yale University) SRCOS 9 / 61

12 Disorders, Genes and Covariates Covariates: interact or confound genetic effects Failure to account for covariates: bias or reduced power Heping Zhang (C 2 S 2, Yale University) SRCOS 10 / 61

13 Study Design Population-based studies Family-based studies Heping Zhang (C 2 S 2, Yale University) SRCOS 11 / 61

14 Outline 1 Background Comorbidity Disorders, Genes and Covariates 2 Weighted Association Test Generalized Kendall s Tau Asymptotic Distribution and Power Application to WTCCC Bipolar Disorder Data 3 Maximum Weighted Association Test Asymptotic Distribution Simulations Application to COGA Data 4 Conclusions Heping Zhang (C 2 S 2, Yale University) SRCOS 12 / 61

15 Notation and Hypothesis n study subjects, from a population-based study or family-based study For each subject: A vector of traits T = (T (1),..., T (p) ) Marker genotype M Parental marker genotypes M pa (only available in a family-based study) A vector of covariates Z = (Z (1),..., Z (l) ) Null hypothesis: no association between marker alleles and any linked locus that influences traits T Heping Zhang (C 2 S 2, Yale University) SRCOS 13 / 61

16 Typical Response Fagerstrom Test for Nicotine Dependence Heping Zhang (C 2 S 2, Yale University) SRCOS 14 / 61

17 Multivariate Distributions t n t! a n t,a! a ˆp nt,a t,a exp { 1 2 (x µ) Σ 1 (x µ)} (2π) n Σ Heping Zhang (C 2 S 2, Yale University) SRCOS 15 / 61

18 Kendall s Tau A nonparametric statistic measuring the rank correlation between two variables Pairs of observations: {(X i, Y i ) : i = 1,..., n} (X i, Y i ) and (X j, Y j ): Concordant, if X i X j and Y i Y j have the same sign Disconcordant, if X i X j and Y i Y j have the different sign Kendall s tau: τ = 2(A B)/{n(n 1)} A and B: numbers of concordant and disconcordant pairs Or τ = ( ) n 1 sign{(x i X j )(Y i Y j )} 2 i<j Heping Zhang (C 2 S 2, Yale University) SRCOS 16 / 61

19 Generalized Kendall s Tau F ij = {f 1 (T (1) i T (1) j ),..., f p (T (p) i T (p) j )} f k ( ): identity function for a quantitative or binary trait f k ( ): sign function for an ordinal trait D ij = C i C j. C: number of any chosen allele in marker genotype M Genaralized Kendall s tau (Zhang, Liu and Wang, 2010): U = ( ) n 1 D ij F ij 2 i<j Special cases: FBAT-GEE (Lange et al. 2003) Test for a single ordinal trait (Wang, Ye and Zhang, 2006) Heping Zhang (C 2 S 2, Yale University) SRCOS 17 / 61

20 Outline 1 Background Comorbidity Disorders, Genes and Covariates 2 Weighted Association Test Generalized Kendall s Tau Asymptotic Distribution and Power Application to WTCCC Bipolar Disorder Data 3 Maximum Weighted Association Test Asymptotic Distribution Simulations Application to COGA Data 4 Conclusions Heping Zhang (C 2 S 2, Yale University) SRCOS 18 / 61

21 Hypothesis with Covariates New null hypothesis: no association between marker alleles and any linked locus that influences traits T conditional on covariates Z Heping Zhang (C 2 S 2, Yale University) SRCOS 19 / 61

22 Weighted Test A weight function w(z i, Z j ) imposes a relatively large weight when Z i is close to Z j, and a relatively small weight when Z i and Z j are far away Weighted U-statistic: S = Weighted test statistic: ( ) n 1 D ij F ij w(z i, Z j ) 2 i<j χ 2 τ = {S E 0 (S)} Var 1 0 (S){S E 0(S)} Heping Zhang (C 2 S 2, Yale University) SRCOS 20 / 61

23 Weight Function I: Distance Write Z = (Z co, Z ca ), with Z co for the continuous covariates and Z ca for the categorical covariates For example, w(z i, Z j ; h, q) = W h ( Z co i Z co j )W q {I(Z ca i Z ca j )} W h (u) = exp( u 2 /2h 2 ), h > 0, W q (v) = (1 q)i(v = 0) + qi(v = 1), 0 q 0.5 Weighted U-statistic (called fixed-(h, q) U-statistic): S(h, q) = ( ) n 1 D ij F ij w(z i, Z j ; h, q) 2 i<j Heping Zhang (C 2 S 2, Yale University) SRCOS 21 / 61

24 Weight Function II: Propensity Score Propensity score: probability of a unit being assigned to a particular treatment given a set of covariates Causal effect analysis: match subjects according to their propensity scores (Rosenbaum and Rubin, 1984) Genomic propensity score: p(z) = {p 1 (z), p 2 (z)}, p c (z) = P(C = c Z = z) Genetic association analysis: match subjects according to their genomic propensity scores Weight function: w(z i, Z j ) = W h { p(z i ) p(z j ) }, with W h (u) = exp( u 2 /2h 2 ), h > 0 Heping Zhang (C 2 S 2, Yale University) SRCOS 22 / 61

25 Asymptotic Distribution: Setting Treating the offspring genotype as random Conditioning on all phenotypes and parental genotypes (if available) Eliminates the assumptions about phenotype distribution, genetic model and parental genotype distribution Robust and less prone to population stratification In addition, conditioning on covariates Heping Zhang (C 2 S 2, Yale University) SRCOS 23 / 61

26 Asymptotic Distribution: Null Hypothesis When n, Var 1/2 D 0 {S(h, q)}[s(h, q) E 0 {S(h, q)}] N(0, I p ) Fixed-(h, q) test statistic: Mean and variance: χ 2 τ (h, q) D χ 2 p E 0 {S(h, q)} = Var 0 {S(h, q)} = 2 n 1 n i=1 4 (n 1) 2 ū i E 0 (C i M pa i, Z i ), n i=1 n i=1 ū i ū jcov 0 (C i, C j M pa i, M pa j, Z i, Z j ), with ū i = n 1 n j=1 F ijw(z i, Z j ; h, q) Heping Zhang (C 2 S 2, Yale University) SRCOS 24 / 61

27 Asymptotic Distribution: Power Under the alternative hypothesis, χ 2 τ (h, q) p e i χ 2 1(φ i ) i=1 µ = µ 1 µ 0 E 1 {S(h, q)} E 0 {S(h, q)} Σ 0 = Var 0 {S(h, q)} Σ 1 = Var 1 {S(h, q)} e 1 e p 0: eigenvalues of Σ 1/2 1 Σ 1 0 Σ1/2 1 φ i = µ 2 i µ i : ith component of µ = QΣ 1/2 1 µ Q: an orthonormal matrix, QΣ 1/2 1 Σ 1 0 Σ1/2 1 Q = diag(e 1,..., e p ) Heping Zhang (C 2 S 2, Yale University) SRCOS 25 / 61

28 Factors Determining the Power The conditional power P: P = P { p i=1 e iχ 2 1 (φ i) q χ 2 p (1 α) } Taking a family-based study as an example, µ 1 = Σ 1 = 2 n 1 n i=1 4 (n 1) 2 ū i E(C i T i, Z i, M pa i ) n i=1 n j=1 ū i ū jcov(c i, C j T i, T j, Z i, Z j, M pa i, M pa j ) By Bayes theorem, P(C = c T, Z, M pa ) = P(T C=c,Z)P(C=c Mpa ) c P(T C=c,Z)P(C=c M pa ) Penetrance: P(T C = c, Z) Allele frequency: P(C = c M pa ) Heping Zhang (C 2 S 2, Yale University) SRCOS 26 / 61

29 Power Approximation Using the result from Liu et al. (2009), we have Theorem P P{χ 2 l (ν) q }, where l, ν, and q depend on µ 1 and Σ 1. Heping Zhang (C 2 S 2, Yale University) SRCOS 27 / 61

30 Outline 1 Background Comorbidity Disorders, Genes and Covariates 2 Weighted Association Test Generalized Kendall s Tau Asymptotic Distribution and Power Application to WTCCC Bipolar Disorder Data 3 Maximum Weighted Association Test Asymptotic Distribution Simulations Application to COGA Data 4 Conclusions Heping Zhang (C 2 S 2, Yale University) SRCOS 28 / 61

31 WTCCC Bipolar Disorder Data Collected by Wellcome Trust Case-Control Consortium (WTCCC, 2007, Nature) Phenotype: 1998 cases/3004 controls of bipolar disorder Genotype: genotyped by Affymetrix GeneChip 500K arrays Covariates: gender, age at recruitment Our method: weighted test using propensity score approach (h = 1) Methods for comparison: non-weighted test and logistic regression Strong association: p-value < ; moderate association: < p-value < 10 5 Heping Zhang (C 2 S 2, Yale University) SRCOS 29 / 61

32 Manhattan Plot: Comparison of Three Methods Heping Zhang (C 2 S 2, Yale University) SRCOS 30 / 61

33 GWAS Results Chr. SNP Position Non-weighted Weighted Logistic Regression 6 rs e e e-9 16 rs e e e-9 16 rs e e e-6 16 rs e e e-6 16 rs e e e-6 17 rs e e e-7 20 rs e e e-7 20 rs e e e-7 Heping Zhang (C 2 S 2, Yale University) SRCOS 31 / 61

34 Propensity Scores: Estimation Chr. SNP Gender Age Coefficient p-value Coefficient p-value 16 rs rs rs Heping Zhang (C 2 S 2, Yale University) SRCOS 32 / 61

35 Propensity Scores: Histograms rs : histogram of propensity score 1 rs : histogram of propensity score 2 Frequency Frequency score[, 1] score[, 2] rs : histogram of propensity score 1 rs : histogram of propensity score 2 Frequency Frequency score[, 1] score[, 2] rs : histogram of propensity score 1 rs : histogram of propensity score 2 Frequency Frequency score[, 1] score[, 2] Heping Zhang (C 2 S 2, Yale University) SRCOS 33 / 61

36 Significant Region 29kb region: 7 strongly linked SNPs Haplotype association p-value: by weighted test Heping Zhang (C 2 S 2, Yale University) SRCOS 34 / 61

37 Interactions Between rs and New SNPs Model Variable Coefficient p-value rs e-8 (1) rs e-8 rs rs rs e-3 (2) rs e-5 rs rs rs e-3 (3) rs e-5 rs rs rs e-4 (4) rs e-5 rs rs Heping Zhang (C 2 S 2, Yale University) SRCOS 35 / 61

38 Interactions Among New SNPs Model Variable Coefficient p-value rs e-3 (1) rs e-5 rs rs e-3 (2) rs e-5 rs rs e-3 (3) rs e-5 rs rs Heping Zhang (C 2 S 2, Yale University) SRCOS 36 / 61

39 Outline 1 Background Comorbidity Disorders, Genes and Covariates 2 Weighted Association Test Generalized Kendall s Tau Asymptotic Distribution and Power Application to WTCCC Bipolar Disorder Data 3 Maximum Weighted Association Test Asymptotic Distribution Simulations Application to COGA Data 4 Conclusions Heping Zhang (C 2 S 2, Yale University) SRCOS 37 / 61

40 Maximum Weighted Test Fixed-(h, q) test: how to choose optimal parameters h and q? Choose a grid of h and q values and maximize the weighted test statistic over those choices {h 1,..., h L1 }: pre-specified grid points of h {q 1,..., q L2 }: pre-specified grid points of q χ 2 τ,max = max 1 l 1 L 1,1 l 2 L 2 χ 2 τ (h l1, q l2 ) Approximate the optimal weighting scheme, yielding the strongest association measure Heping Zhang (C 2 S 2, Yale University) SRCOS 38 / 61

41 Resampling Approach Population-based studies: restricted permutation in Yu et al. (2010) Family-based studies: children s genotypes solely determined by their parents marker alleles, resample the children s genotype by Mendelian laws Calculate M resampling test statistics χ 2 τ,max,1,..., χ2 τ,max,m using M resampled data Resampling p-value: the proportion of the resampling test statistics that exceed our observed test statistic, i.e., M 1 M m=1 I( χ2 τ,max,m χ 2 τ,max) Heping Zhang (C 2 S 2, Yale University) SRCOS 39 / 61

42 Asymptotic Distribution: Joint Distribution Equivalently, χ 2 τ,max = max R l1,l l 1 L 1,1 l 2 L 2 R = Var 1/2 0D (S){S E 0(S)} S = {S (h 1, q 1 ),..., S (h L1, q L2 )} Var 0D (S) = diag[var 0 {S(h 1, q 1 )},..., Var 0 {S(h L1, q L2 )}]: the diagonal blocks of Var 0 (S) Var 1/2 0 (S){S E 0 (S)} D N(0, I pl1 L 2 ) R = Var 1/2 0D (S)Var1/2 0 (S)G, G N(0, I pl1 L 2 ) Heping Zhang (C 2 S 2, Yale University) SRCOS 40 / 61

43 Asymptotic Distribution: Uniform Approximation Theorem Assume that the eigenvalues of Var 0D (S) and Var 0 (S) are uniformly bounded from both above and below, i.e., there exist two positive numbers c and C such that c λ min {Var 0D (S)} λ max {Var 0D (S)} C and c λ min {Var 0 (S)} λ max {Var 0 (S)} C uniformly for all n, where λ min and λ max denote the smallest and largest eigenvalues respectively. Then for any x R, as n, ( ) ( ) sup P χ 2 τ,max x P max R l1,l 2 2 x 0. 1 l 1 L 1,1 l 2 L 2 x R Heping Zhang (C 2 S 2, Yale University) SRCOS 41 / 61

44 Outline 1 Background Comorbidity Disorders, Genes and Covariates 2 Weighted Association Test Generalized Kendall s Tau Asymptotic Distribution and Power Application to WTCCC Bipolar Disorder Data 3 Maximum Weighted Association Test Asymptotic Distribution Simulations Application to COGA Data 4 Conclusions Heping Zhang (C 2 S 2, Yale University) SRCOS 42 / 61

45 Aims Compare the performance of: Maximum weighted test Non-weighted test Compare the performance of: Maximum weighted test Other covariate-adjusted tests Heping Zhang (C 2 S 2, Yale University) SRCOS 43 / 61

46 Simulation I: Data Generation Generate the parents disease and marker genotypes via the haplotype frequencies Given the parental genotypes, generate the offspring genotype using 1cM between the two loci Two covariates are generated: Z co N(1, 2) Without confounder: P(Z ca = 1) = 1 P(Z ca = 0) = 0.7 With confounder: logit{p(z ca = 1)} = 0.5M fa + 0.5M mo Bivariate ordinal traits are generated according to random effects proportional odds model: logit{p(t (j) k)} = α j,k + β g G + β co Z co + β ca Z ca + U j, with k = 1,..., K j, j = 1, 2, and (U 1, U 2 ) N(0, Σ) Heping Zhang (C 2 S 2, Yale University) SRCOS 44 / 61

47 Simulation I: Parameters Number of categories: K 1 = 3 and K 2 = 4 (α 1,1, α 1,2 ) = ( 0.5, 0.3), (α 2,1, α 2,2, α 2,3 ) = ( 0.5, 0.3, 0.1) β g = 2.0 β co = β ca = 0, 0.5, 1.0, 1.5, 2.0 ( ) Σ = The grid of h is {C 1 (C 2 /C 1 ) (l 1/(L 1 1)) : l 1 = 0,..., L 1 1}, with C 1 = 0.05, C 2 = 10, L 1 = 8 The grid of q is {0.5l 2 /(L 2 1) : l 2 = 0,..., L 2 1}, with L 2 = 5 Heping Zhang (C 2 S 2, Yale University) SRCOS 45 / 61

48 Type I Error of Maximum Weighted Test Significance Level Confounder n α = 0.05 α = 0.01 α = No Yes Heping Zhang (C 2 S 2, Yale University) SRCOS 46 / 61

49 Power Comparison: Without Confounder Covariate effect n α Method Weighted Non-weighted Weighted Non-weighted Weighted Non-weighted Weighted Non-weighted Weighted Non-weighted Weighted Non-weighted Heping Zhang (C 2 S 2, Yale University) SRCOS 47 / 61

50 Power Comparison: With Confounder Covariate effect n α Method Weighted Non-weighted Weighted Non-weighted Weighted Non-weighted Weighted Non-weighted Weighted Non-weighted Weighted Non-weighted Heping Zhang (C 2 S 2, Yale University) SRCOS 48 / 61

51 Simulation II: Other Covariate-Adjusted Methods FBAT-GEE (Lange et al. 2003) adjusting for covariates: Fit the regression model g(e[t (j) ]) = α j + λ jz, with g( ) an appropriate link function Replace the original traits T (j) with the residuals T (j) g 1 (α j + λ jz) in the FBAT-GEE test statistic Ordinal trait test (Wang, Ye and Zhang, 2006): Deal with a single ordinal trait at a time Apply the Bonferroni correction for multiple trait testing Heping Zhang (C 2 S 2, Yale University) SRCOS 49 / 61

52 Simulation II: Data Generation Continuous covariate: Z co N(1, 2) Bivariate quantitative traits: Y (j) = µ + β g G + β co Z co + ɛ j, j = 1, 2, with (ɛ 1, ɛ 2 ) 2φ 2 (x; Σ)Φ(α x), x R 2 (bivariate skew normal distribution) Bivariate ordinal traits: discretizing quantitative traits by (50%, 67%) and (33%, 54%, 75%) percentiles ( ) α = (5, 5) and Σ = µ = 0, β g = β co = 0.8 Heping Zhang (C 2 S 2, Yale University) SRCOS 50 / 61

53 Power Comparison Significance Level n Method α = 0.05 α = 0.01 α = χ 2 τ,max FBAT-GEE Wang et al. s Test χ 2 τ,max FBAT-GEE Wang et al. s Test χ 2 τ,max FBAT-GEE Wang et al. s Test Heping Zhang (C 2 S 2, Yale University) SRCOS 51 / 61

54 Outline 1 Background Comorbidity Disorders, Genes and Covariates 2 Weighted Association Test Generalized Kendall s Tau Asymptotic Distribution and Power Application to WTCCC Bipolar Disorder Data 3 Maximum Weighted Association Test Asymptotic Distribution Simulations Application to COGA Data 4 Conclusions Heping Zhang (C 2 S 2, Yale University) SRCOS 52 / 61

55 Collaborative Studies on Genetics of Alcoholism A large scale study to map alcohol dependence susceptible genes Heping Zhang (C 2 S 2, Yale University) SRCOS 53 / 61

56 COGA Data The data include 143 families with a total of 1,614 individuals Multiple Traits: ALDX1 (the severity of the alcohol dependence): pure unaffected, never drunk, unaffected with some symptoms, and affected MaxDrink (maximum number of drinks in a 24 hour period): 0-9, 10-19, 20-29, and more than 30 drinks TimeDrink (spent so much time drinking, had little time for anything else): no, yes and lasted less than a month, and yes and lasted for one month or longer Genotypes: markers on chromosome 7 Covariates: age at interview and gender Heping Zhang (C 2 S 2, Yale University) SRCOS 54 / 61

57 Results D7S1790 D7S513 D7S1802 D7S629 D7S673 D7S1838 NPY2 D7S817 D7S2846 D7S521 D7S691 D7S478 D7S679 D7S665 D7S1830 D7S3046 D7S1870 D7S1797 D7S820 D7S821 D7S1796 D7S1799 D7S1817 D7S2847 D7S490 D7S1809 log(p value) D7S1804 D7S509 D7S1824 D7S794 D7S1805 adjusted unadjusted Distance (cm) Heping Zhang (C 2 S 2, Yale University) SRCOS 55 / 61

58 Conclusions: Method Developed a nonparametric weighted test to adjust for covariates that accommodates multiple traits Provided its asymptotic distribution and analytical power calculation Refined the weighted test by proposing the idea of maximum weighting over the grid points of parameters Proposed an asymptotic approach to assess its significance Heping Zhang (C 2 S 2, Yale University) SRCOS 56 / 61

59 Conclusions: Application WTCCC bipolar disorder data: not only confirmed the results reported by the WTCCC (2007), but also identified another region at the genome-wide significance level The identified haplotype block is near the RPGRIP1L gene that was reported to be associated with bipolar disorder (O Donovan et al., 2008; Riley et al., 2009) COGA data: confirmed and strengthened the top signal; provided evidences for the advantage of maximum weighted test over non-weighted test Heping Zhang (C 2 S 2, Yale University) SRCOS 57 / 61

60 Other Ongoing/Future Work Incorporating genetic prior information into a current study Genetic association analysis for rare variants Nonparametric test for gene-environment interactions Genetic test for multiple trait covariance structure Heping Zhang (C 2 S 2, Yale University) SRCOS 58 / 61

61 Acknowledgment Dr. Yuan Jiang, Yale University Dr. Ching-Ti Liu, Boston University Dr. Xueqin Wang, Sun Yat-Sen University, China Dr. Wensheng Zhu, Northeastern Normal University, China Heping Zhang (C 2 S 2, Yale University) SRCOS 59 / 61

62 References W. Zhu and H. Zhang (2009) Why do we test multiple traits in genetic association studies? (with discussion). J. Korean Statist. Soc., 38:1-10. H. Zhang, C.-T. Liu and X. Wang (2010) An association test for multiple traits based on the generalized Kendall s tau. J. Amer. Statist. Assoc., 105: Y. Jiang and H. Zhang (2011) Propensity Score-Based Nonparametric Test Revealing Genetic Variants Underlying Bipolar Disorder. Genet. Epidemiol., 35: H. Zhang (2011) Statistical Analysis in Genetic Studies of Mental Illnesses, Statistical Science, in press. W. Zhu, Y. Jiang, and H. Zhang (2011) Covariate-Adjusted Association Tests and Power Calculations Based on the Generalized Kendall s Tau. Revised and under review. Heping Zhang (C 2 S 2, Yale University) SRCOS 60 / 61

63 Heping Zhang (C 2 S 2, Yale University) SRCOS 61 / 61

Genetic Studies of Multivariate Traits

Genetic Studies of Multivariate Traits Heping Zhang Collaborative Center for Statistics in Science Yale University School of Public Health Presented at DIMACS, University of Rutgers May 17, 2013 Heping