Integrating GWAS with omic dataweip/paper/CCBR19.pdf · 2019-03-01 · Integrating GWAS with omic...

27
Integrating GWAS with omic data Wei Pan Division of Biostatistics, School of Public Health, University of Minnesota, Minneapolis, MN 55455 Joint work with Zhiyuan Xu (LinkedIn), Chong Wu (FSU), Haoran Xue, ... March 1, 2019 CCBR

Transcript of Integrating GWAS with omic dataweip/paper/CCBR19.pdf · 2019-03-01 · Integrating GWAS with omic...

Integrating GWAS with omic data

Wei Pan

Division of Biostatistics, School of Public Health, University of Minnesota,Minneapolis, MN 55455

Joint work with Zhiyuan Xu (LinkedIn), Chong Wu (FSU), Haoran Xue, ...

March 1, 2019CCBR

Outline

I Introduction: problem

I Review: Transcriptome-Wide Association Study(TWAS)/PrediXcan

I Our method: (weighted) aSPU test.

I GWAS + gene expression: application to the Lipid data

I GWAS + EPI + Methylation: application to the SCZ data

I On-going: Mendelian randomization (MR)determining causal direction between two variables: LDL/HDLvs CAD.

Introduction

I Problem: to detect SNP-disease associations in GWAS (orsequencing studies).Big question: mechanistic interpretation?

I Approach: integrating GWAS data with eQTL data1) to boost power;2) to enhance interpretation.

I Motivation: DNA =⇒ mRNA =⇒ ...=⇒ Disease.Known: many disease-associated SNPs are eQTL.

TWAS/PrediXcan: GWAS + eQTL

I Set-up: 2 independent data sets, one GWAS (main), oneeQTL (smaller).GWAS: SNPs, Y (disease);eQTL: SNPs, X (gene expression).

I Conventional GWAS: Y ∼ SNP.

I PrediXcan (Gamazon et al 2015, Nat Genet), TWAS (Gusevet al (2016, Nat Genet):1. eQTL: E (X ) = SNPs′ ∗ w =⇒ w ;2. GWAS: 1) X = SNPs ′ ∗ w ; 2) Y ∼ X

I Why? why not Y ∼ X?Biologically: genetically regulated component of geneexpression (GReX);X not available in the GWAS data, often.

Our views

I Statistically: (two-sample) 2SLSrelated to Mendelian randomization (MR); causal inference!

I Our key obs: PrediXcan/TWAS = weighted SPU(1)!weight each SNP j by wj , then ...

I Why not use aSPU (or other more powerful tests)? (Xu et al2017, Genetics; Su et al 2018, AJHG)

I Why not use other weights derived from other omic orendophenotypes?brain imaging for AD (Xu et al 2017, NeuroImage);enhancer-promoter interactions (EPIs) (Wu and Pan 2018,Genetics);EPIs + meQTL (Wu and Pan 2019, BI).

Lipid GWAS + eQTL

I Discovery data: a large 2010 Lipid dataset (Teslovich et al2010, Nat Genet).Summary stats of meta-analysis of 46 GWAS withn ≈ 100, 000 individuals;Lipid traits: LDL, HDL, TG, TC; use LDL here.

I Validation data: a larger 2013 dataset (GLGC 2013, NatGenet).n = 188, 577.

I Three eQTL datasets: NTR, YFS and METSIM;extracted the (optimal) weights constructed by Gusev et al(2016);containing 1264, 3555 and 2295 (and 1223 for a combinedanalysis) genes.

I As in Gusev et al (2016), for gene-based analysis,conservatively usegenome-wide significance level=0.05/8500=5.88E-6.

I Applied our aSPU and TWAS=SPU(1).

Table: The numbers of the significant genes identified by analyzing the2010 lipid data. a/b/c indicate the numbers of (a) the significant genes;(b) the significant genes that covered a genome-wide significant SNPs inthe 2010 lipid data; (c) the significant genes that covered a genome-widesignificant SNPs in the 2013 lipid data.

Trait Test NTR YFS METSIM Combined

HDL aSPU 19/16/17 29/27/29 22/19/22 21/17/17TWAS 16/14/15 25/22/24 19/15/19 20/16/17

LDL aSPU 15/15/15 19/18/18 17/16/17 14/13/13TWAS 8/7/8 10/9/9 7/7/7 7/7/7

TG aSPU 17/16/17 33/30/32 15/14/14 20/19/19TWAS 9/9/9 17/16/17 8/7/7 12/11/11

TC aSPU 26/25/26 28/26/27 28/28/28 20/20/20TWAS 15/14/15 18/16/17 15/14/15 14/13/13

0

2

4

6

8

10

Chromosome

−lo

g 10(p

)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 20 22

●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●●●●●

●●●●

●●

●●

●●●●●

●●

●●

●●

●●●

●●

●●●●

●●

●●●●

●●●

●●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●●

●●

●●●●●

●●●

●●

●●

●●

●●

●●

●●

●●●●

●●●●●●

●●●

●●

●●

●●

●●

●●

●●

●●●●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●●●●●●●●

●●

●●●●

●●

●●●●

●●●

●●●

●●●

●●

●●●

●●●●

●●

●●

●●●

●●

●●●

●●

●●

●●●

●●

●●●●●●

●●

●●●●

●●

●●

●●

●●

●●●

●●●●

●●

●●●

●●

●●

●●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●●●

●●

●●●●

●●

●●

●●●●

●●

●●

●●

●●●●

●●

●●●

●●●●●●●

●●

●●●

●●●

●●●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●●●

●●●

●●

●●●

●●

●●●●

●●●●

●●

●●●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●●

●●

●●●

●●

●●●●●●

●●

●●

●●●●●●●●

●●●●●

●●

●●

●●

●●●●●

●●●

●●●●●

●●

●●●

●●

●●●

●●

●●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●●

●●●

●●●

●●●●

●●

●●

●●

●●●●

●●●

●●

●●●●●●

●●

●●●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●●●●

●●

●●●

●●●●●●●●●●

●●

●●●

●●●

●●●●

●●●●●●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●●●●●●

●●

●●

●●●

●●

●●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●●●

●●

●●

●●●

●●●

●●

●●●●●

●●●

●●

●●●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●●●●●●

●●

●●

●●

●●

●●

●●

●●●●●

●●

●●

●●●●

●●●

●●

●●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●●

●●●

●●●

●●

●●

●●

●●●

●●

●●

●●●

●●●

●●●●

●●

●●●●

●●

●●

●●

●●

●●●●●●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●●●

●●

●●

●●

●●

●●●

●●

●●●

●●●

●●●

●●

●●●

●●●●●

●●●●●

●●●●

●●

●●●

●●●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●●

●●●

●●

●●●●

●●●●

●●●●

●●

●●

●●●

●●●●

●●

●●

●●●●●

●●●

●●

●●

●●●●●

●●

●●

●●

●●

●●

●●●

●●●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●●●●●●●●●

●●

●●

●●●●●

●●

●●

●●

●●●

●●

●●●

●●●

●●●●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●●●

●●

●●●

●●●●●

●●

●●

●●

●●●●

●●

●●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●●●

●●●

●●

●●

●●●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●

●●

●●●●●●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●●●●●

●●

●●

●●

●●●

●●

●●

●●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●●●

●●

●●●●

●●

●●●●●

●●

●●●●

●●●

●●

●●

●●

●●●

●●●

●●

●●

●●●

●●

●●●

●●

●●●●

●●

●●●

●●●●●●

●●

●●●

●●

●●

●●●●

●●●

●●

●●

●●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●●

●●

●●●●●●

●●●●●

●●

●●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●●

●●

●●●

●●

●●

●●●

●●●

●●●

●●

●●

●●●●

●●

●●

●●●

●●

●●●

●●●●

●●●●●●●

●●

●●●●●●

●●

●●●●

●●

●●

●●

●●●●

●●●●●

●●

●●●

●●

●●●●●

●●

●●

●●●●●

●●●●

●●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●●●

●●

●●●

●●

●●

●●●●●●

●●

●●

●●

●●

●●●●●●●●

●●●●

●●●●

●●

●●

●●

●●

●●

●●

●●●●●

●●

●●

●●

●●

●●●

●●●●●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●●●●●

●●

●●●

●●

●●

●●●●●●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●●●

●●●●●●

●●●

●●

●●●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●●●●●

●●●●●●●●

●●

●●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●●

●●

●●●

●●●●

●●

●●

●●

●●

●●●●●

●●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●●

●●

●●●●

●●

●●

●●

●●

●●●●●●●●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●●

●●

●●

●●●

●●●

●●●

●●●

●●●●

●●●

●●

●●

●●●●●●●●

●●●

●●●●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●●●●

●●

●●

●●

●●●●

●●

●●●

●●

●●

●●

●●●

●●●●

●●

●●

●●●●

●●●

●●●●

●●

●●

●●●●

●●

●●●

●●●

●●●

●●

●●●

●●

●●

●●●

●●●

●●

●●●●●●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●●●●●

●●

●●

●●

●●

●●

●●●●

●●

●●●●●

●●●●

●●

●●

●●●

●●

●●●

●●

●●

●●●

●●●●

●●●

●●

●●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●●

●●●●

●●

●●

●●

●●

●●●●

●●

●●●

●●

●●●

●●●

●●●

●●●●

●●●●●●

●●●

●●

●●

●●●

●●●●

●●

●●

●●

●●

●●

●●

●●●

●●●●

●●

●●●●●●

●●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●●●

●●

●●●●●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●●

●●

●●●

●●●●●●●

●●

●●

●●

●●●

●●●●

●●

●●

●●●●

●●●●

●●●

●●●

●●

●●

●●

●●●●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●●●●

●●

●●

●●●●

●●

●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●●●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●●●

●●●●

●●

●●●●

●●

●●●

●●

●●●●●●●

●●●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●●

●●

●●●●

●●

●●

●●●

●●

●●●

●●

●●

●●●●●●●●●

●●●●●

●●●

●●●

●●●

●●

●●●

●●●

●●

●●●●

●●

●●

●●●

●●

●●

●●

●●

●●●●●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●●●●

●●●●●

●●

●●●●●●

●●●●●

●●●

●●●●●●

●●

●●

●●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●●

●●●

●●●●●

●●

●●●●

●●●●●

●●

●●●●●●

●●●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●●●

●●

●●

●●●●

●●●

●●●

●●

●●●

●●

●●●●●●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●●●

●●●●

●●

●●●●

●●●●

●●●

●●

●●●●

●●

●●

●●●●

●●●

●●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●●●●●

●●

●●

●●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●●

●●●

●●

●●

●●●●●

●●

●●

●●

●●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●●●

●●●

●●●●

●●

●●●

●●●

●●

●●

●●●●●●

●●●

●●

●●

●●

●●●

●●

●●●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●●●

●●

●●

●●

●●●●

●●

●●●●●

●●●●

●●

●●

●●●●

●●

●●

●●●●

●●

●●●

●●

●●

●●●

●●●●

●●●

●●●●

●●

●●

●●●

●●

●●●

●●●●●

●●

●●●

●●●●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●●

●●●

●●●

●●●

●●

●●●●

●●

●●

HMGN2(m)ARID1A(n)

ZDHHC18(y)PABPC4(y)

DR1(y)DR1(o)PSRC1(y)PSRC1(o)

RBM6(y)RBM6(m)ITIH4(o)RFT1(m)

AFF1(m) FCHO2(o) IER3(n)IER3(o)

SNRPC(y)SNRPC(m)SNRPC(o)

UHRF1BP1(n)UHRF1BP1(y)UHRF1BP1(m)

MLXIPL(m)AOC1(m)

ERI1(n)CSGALNACT1(o)CSGALNACT1(m)

LPL(y)LPL(n)LPL(o)

TTC39B(o)TTC39B(y)

PTPRE(y)ARHGAP1(m)

ACP2(m)NR1H3(y)

MYBPC3(m)SPI1(y)

RAPSN(m)C1QTNF4(m)

MTCH2(n)PTPRJ(n)C11orf10(n)FADS1(y)CTSW(o)

SIDT2(y)SIDT2(m)SIDT2(o)TAGLN(y)PCSK7(m)PCSK7(o)

KCTD10(o)UBE3B(n)UBE3B(o)ALDH2(o)ARL6IP4(y)

CDK2AP1(n)CDK2AP1(o)CDK2AP1(y)

SETD8(y)SNRNP35(y)

EIF2B1(n)EIF2B1(o)TCTN2(m)

CCDC92(m)CCDC92(o)NCOR2(y)SCARB1(y)BRI3BP(y)

LIPC(n) BBS2(n)BBS2(o)MT2A(y)MT1E(m)MT1E(o)

HERPUD1(y)NLRC5(y)NLRC5(n)NLRC5(o)COQ9(y)ELMO3(m)ELMO3(o)

ATP6V0D1(m)ATP6V0D1(o)

LCAT(y)DPEP3(n)DPEP2(y)DUS2L(y)DUS2(m)

NFATC3(y)ESRP2(n)SLC7A6(o)SLC7A6(y)PRMT7(y)PRMT7(m)PRMT7(o)

FBXL20(y)PGAP3(y)PGAP3(m)PGAP3(o)IKZF3(n)

GSDMB(n)ORMDL3(n)

BLOC1S3(m)LILRA3(n)LILRA3(o)LILRA3(y)LILRA5(y)

PLTP(m)MMP9(y)

UBE2L3(y)CCDC116(m)

TMED5(o)

CYB561D1(y)

GSTM3(o)

GSTM3(m)

SLC41A1(y)

RETSAT(n) RETSAT(o)

RETSAT(y)

RBM6(n)

STAB1(y)

ITIH4(y)

TMEM110(m)

SNX10(m)

AGAP4(m)

KCTD10(y)KCTD10(m)

ALDH2(y)

VPS37B(y)

SBNO1(y)

ELMO3(y)

GSDMB(y)ORMDL3(y)

GSDMA(m)

GSDMA(o)FPR3(y)

ASCC2(y)

Figure: Manhattan plots for the pooled results of aSPU and aSPU-Ofortraits HDL based on the 2013 lipid data.

0

2

4

6

8

10

Chromosome

−lo

g 10(p

)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 20 22

●●

●●●

●●

●●

●●●

●●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●●●

●●●

●●●

●●

●●

●●

●●

●●●●●

●●●●

●●

●●

●●

●●●

●●●●●●

●●

●●●

●●

●●●

●●

●●●

●●●●

●●

●●

●●●●

●●

●●●

●●

●●

●●

●●

●●●●●●●

●●

●●

●●●●●●

●●●●●●

●●

●●●

●●

●●

●●

●●●

●●●●●

●●

●●

●●

●●

●●●

●●●

●●

●●●●

●●●●●●●●●●●●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●●●●

●●

●●

●●●●●●

●●

●●●

●●

●●●

●●

●●

●●

●●●●

●●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●●

●●●

●●●

●●

●●

●●

●●●

●●

●●

●●●

●●●●●●●●

●●

●●●

●●●●●

●●

●●

●●●

●●●●●●●

●●

●●●

●●●

●●●

●●●

●●●●

●●●●

●●●●●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●●●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●●

●●

●●●

●●●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●●

●●●

●●●

●●●

●●

●●

●●●

●●●●

●●

●●●

●●

●●

●●

●●

●●●●●●

●●

●●

●●

●●●●●

●●●●●●●●

●●●●●●●●●●●●●●

●●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●●●

●●

●●

●●●●●●●●●●●

●●●

●●

●●

●●

●●

●●●

●●

●●●●●

●●

●●

●●●

●●

●●

●●●●

●●

●●

●●

●●●

●●

●●●

●●●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●●

●●●●

●●●●

●●●●

●●●●

●●

●●

●●

●●●

●●

●●●●●

●●

●●

●●

●●

●●●●

●●●●

●●●

●●●●●

●●

●●

●●

●●

●●●

●●

●●●

●●●●●

●●

●●

●●

●●●●●

●●

●●

●●

●●●

●●

●●

●●●●●●●●

●●

●●●

●●

●●

●●●

●●●

●●

●●●

●●

●●●●

●●

●●

●●

●●

●●

●●●●●●●

●●

●●

●●●

●●●

●●●

●●

●●

●●

●●

●●●●●●●

●●●

●●

●●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●●●●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●●●

●●

●●●

●●

●●●

●●●●

●●

●●

●●

●●●●●

●●

●●●●

●●●●●

●●●●

●●●●

●●●●●

●●

●●●●

●●●

●●

●●

●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●●

●●●●●●●●●

●●

●●●●

●●●●●

●●

●●

●●

●●

●●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●●

●●●

●●

●●

●●●●

●●

●●●

●●●

●●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●●

●●●

●●●●●●●

●●

●●

●●

●●

●●●●●

●●●●

●●●

●●

●●●●●

●●

●●●●●●●●●●●

●●

●●●●

●●

●●

●●

●●●

●●●●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●●

●●●●

●●●

●●

●●

●●●●

●●●

●●

●●

●●●

●●●

●●●

●●●●

●●

●●

●●●

●●●

●●

●●

●●

●●●

●●●●●●

●●

●●●

●●

●●●

●●

●●●●●

●●●

●●●●●●●●

●●

●●

●●

●●

●●●

●●●●

●●

●●

●●

●●●

●●

●●●●

●●

●●●●

●●

●●

●●●

●●●

●●●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●

●●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●●●●

●●●

●●●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●●

●●

●●●

●●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●●●●

●●●●

●●

●●

●●

●●

●●●●●

●●

●●

●●

●●

●●●

●●

●●●●

●●

●●●●●

●●●

●●●●

●●●

●●●

●●

●●●

●●

●●

●●●

●●●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●●

●●

●●●

●●●●●●●●●

●●

●●

●●●●●●●

●●

●●

●●

●●

●●●●●●

●●

●●

●●●●

●●●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●●●●●●●

●●

●●

●●

●●

●●●

●●

●●●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●●●

●●●●●●●●●

●●

●●●●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●●●●

●●

●●

●●●

●●

●●

●●●●

●●

●●

●●●

●●

●●●●

●●

●●

●●●

●●

●●●●●●

●●●

●●●

●●●

●●

●●

●●

●●●●●

●●●●●

●●

●●●

●●

●●●

●●

●●

●●●

●●●●

●●

●●●●●●●

●●●●

●●●●●●

●●●

●●

●●

●●

●●●●

●●●●●

●●●

●●

●●●●●

●●●●●

●●

●●

●●

●●●

●●●●●●●

●●

●●●

●●●

●●

●●

●●

●●●●

●●

●●

●●●

●●●

●●

●●

●●●●

●●

●●

●●●●

●●

●●●

●●

●●●

●●●●

●●

●●●

●●●

●●

●●●

●●●●

●●●

●●●●●●

●●●

●●

●●●●

●●

●●●

●●●●

●●

●●

●●●●●

●●

●●●

●●●

●●

●●

●●

●●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●●

●●●●

●●●

●●

●●●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●●

●●●

●●

●●●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●●●

●●●●●

●●

●●

●●●

●●

●●●

●●

●●●●●●●

●●

●●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●●

●●●●●

●●

●●

●●

●●

●●●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●●

●●

●●

●●

●●

●●●●

●●●●●

●●

●●●

●●●

●●

●●●

●●

●●●

●●

●●

●●

●●●●●●

●●

●●●●

●●●

●●

●●●

●●●●

●●

●●

●●●

●●●●

●●●●●●

●●●

●●●●●

●●●

●●

●●●

●●●

●●●

●●●●●

●●

●●

●●●

●●●

●●

●●●

●●●

●●

●●●

●●●●

●●

●●

●●●●

●●

●●●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●●●

●●

●●●●●●

●●●●●●●

●●●●●

●●

●●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●●

●●●

●●

●●●●

●●

●●●●●●

●●

●●

●●●

●●●●

●●

●●

●●

●●

●●

●●●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●●●

●●

●●●

●●●

●●

●●

●●

●●●●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●●●

●●

●●

●●●

●●

●●●●●●●

●●●●

●●

●●●●

●●●

●●

●●●●●●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●●

●●●●

●●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●●

●●●

●●●●

●●

●●

●●

●●

●●

●●●●●

●●●●●

●●

●●

●●

●●●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●●●

●●●

●●

●●●

●●●

●●

●●●

●●●

●●●

●●

●●●●

●●●●

●●●

●●

●●●●●

●●

●●

●●●●

●●

●●●●

●●●

●●

●●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●

●●

●●●

●●●●●●●●

●●

●●

●●●

●●

●●●

●●●

●●●●

●●●●●

●●●●

●●

●●●●●●●

●●●

●●●

●●

●●

●●

●●

●●●●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●●●●

●●

●●

●●●

●●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●●●

●●●

●●

●●

●●

●●

●●●●

●●●

●●●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●●

●●

●●●

●●●

●●

●●●●

●●

●●

●●●●

●●

●●

●●

●●●

●●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●●●●●●

●●●●

●●

●●

●●●

●●

●●●

●●

●●●

●●

●●

●●●●●●●●●●●●

●●

●●

●●

●●●●

●●

●●

●●●

●●●●●●●●●

●●

●●

●●●

●●

●●●

●●●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●●

●●●●●●●

●●

●●

●●

●●

●●●

●●●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●●●●●●●

●●

●●

●●

●●●●●

●●

●●●

●●●

●●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●●●

●●●

●●

●●●●●●

●●

●●●

●●●

●●

●●●●

●●●●

●●●

●●●

●●●●●

●●

●●●

●●

●●

●●●●

●●●

●●

●●

●●●●

●●

●●

●●●●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●●●●●●

●●

●●

●●●●●

●●

●●

●●

●●●

●●

●●

●●●●

●●●●●●●●●

●●

●●

●●●●

●●●●

●●●

●●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●●

●●

●●●●

●●

●●

●●●●

●●

●●

●●●●

●●

●●

●●●

●●

●●

RHD(m)RHCE(o)RHD(n)

TMEM50A(n)TMEM50A(o)

RHCE(m)ARID1A(n)

DOCK7(y)KIAA1324(n)CELSR2(m)PSRC1(y)PSRC1(m)PSRC1(o)SORT1(n)SYPL2(m)

CYB561D1(y)AMIGO1(m)

GNAI3(y)GSTM4(y)GSTM4(m)GSTM4(n)GSTM4(o)GSTM3(o)

TIMP4(m)MKRN2(o)

PCCB(o) POLK(m)POC5(m)

MICB(o)NOTCH4(y)

TAP2(n)TAP2(o)

FRK(m)NT5DC1(m)NT5DC1(o)

WTAP(m)SLC22A1(m)

NUDCD3(m)NPC1L1(m)DDX56(y)TMED4(y)

ERI1(n) NRBP2(y)PLEC(m)GRINA(n)GRINA(y)PARP10(y)PARP10(m)PARP10(o)

C11orf10(n)FADS1(y)FADS2(n)

PCSK7(o)DCPS(y)

ALDH2(n)ALDH2(o)ALDH2(y)

OASL(y)P2RX7(n)P2RX7(o)

HERPUD1(y)ATXN1L(y)

KIAA0174(n)DHODH(m)

HP(m)HP(o)

DHX38(y)DHX38(m)DHX38(o)

TBKBP1(y)TBKBP1(m)

S1PR5(y)CARM1(y)

SMARCA4(n)TSPAN16(n)

EPOR(y)GATAD2A(n)GATAD2A(o)GATAD2A(y)

LPAR2(o)CEACAM19(m)

PVRL2(n)PVRL2(o)

BLOC1S3(m)NTN5(m)

RHCE(n)TMEM50A(y)

ZDHHC18(y)

MCM6(y)

ICA1L(m)

MKRN2(y)

HLA−B(n)

MICB(y)

BAT1(n)DDAH2(y)SLC44A4(n)

STK19(y)

HLA−DRA(n)

HLA−DQB1(y)

SNRPC(o)

NSMAF(y)

TIRAP(m) SH2B3(y)

ATXN2(y)

MAPKAPK5(y)

P2RX4(y)

PFAS(y)

ERAL1(y)

DHRS13(y)

KANK2(n) PVRL2(y)

Figure: Manhattan plots for the pooled results of aSPU and aSPU-O fortraits LDL based on the 2013 lipid data.

(PGC) SCZ GWAS + EPI + meQTL

3 9 58 2

SPU(2) aSPU

SPU(1)

2 2 35 1

SPU(2) aSPU

SPU(1)

Figure: Left: significant genes; right: sigificant and novel ones for”EPI+meQTL”.

(PGC) SCZ GWAS + EPI + meQTL

9

0

0

0

3

7

0

0

2

32 1

2

5

0

E+G E+G+Mtheyl

TWAS STD

4

0

0

0

1

0

0

0

0

20 1

1

4

0

E+G E+G+Methyl

TWAS STD

Figure: Left: significant genes; right: sigificant and novel ones.Combined SPU(1) and SPU(2).

EPI prediction

I TargetFinder: use epi-genomic features (from ENCODE andRoadmap); boosting (> RF) (Whalen et al 2016, Nat Genet).Cao et al (2017, Nat Genet)

I SPEID: use DNA sequence (Singh et al 2016, bioRxiv); DL(Singh et al 2016, bioRxiv).RNN + CNN

I Ours: use DNA seq with two main contributions:1) only CNN;2) transfer learning: use all cell lines before cell line specificlearning.

Our CNN

Figure: Our CNN architecture for EPI prediction.

Performance

Cell lines

AU

PR

0.0

0.2

0.4

0.6

0.8

1.0

Figure: AUPR for the six cell lines based on SPEID (red/bottom bars),our CNN (yellow/middle bars) and our CNN+transfer learning(organge/top bars).

Orienting the causal relation

I Question: causal direction between X and Y ?X =⇒ Y , or Y =⇒ X?

I Previous applications: gene expression =⇒ LDL (or SCZ)?

I Example: LDL =⇒ CAD?Statins; Mendelian randomization (MR) analyses.

I Example: HDL =⇒ CAD?Failed drug trials; MR analyses: inconclusive.

I Example: Education level =⇒ AD?A Lancet Commission (Livingston et al 2017): possible toprevent about 35% of dementia by controlling nine riskfactors: education to a maximum of age 11-12 years, midlifehypertension, midlife obesity, hearing loss, late-life depression,diabetes, physical inactivity, smoking, and social isolation.

Genetics-based methodsI Using genetic data to strengthen causal inference ... (Pingault

et al 2018, Nat Rev Genet).I Use SNPs as anchors/instrumental variables (IVs) (Schadt et

al 2005, Nat Genet; Chen et al 2007, Genom Biol; ...).SNPs =⇒ ...; not the reverse!SNPs: somewhat randomized.take advantage of many existing large-scale GWAS!

I Mediation analysis:Causal inference test (CIT) (Millstein et al 2009, BMC Genet)Limitations: 1) require data (SNP, X , Y );often have two samples: (SNP, X ), (SNP, Y ).2) less robust to measurement errors.

I MR: Steiger’s test (Hemani et al 2017, PLOS Genet)Theory: If SNP =⇒ X =⇒ Y , then |ρgX | > |ρgY |!Main idea: test their difference!Limitation: based on a single SNP, thus low statisticalefficiency and low robustness! —–our task here!

I Others: Pickrell’s (2016, Nat Genet); bi-directional MR ...

Our method

I Motivation: extending MR Steiger’s method from using asingle SNP to multiple SNPs.1) multiple correlated SNPs in a locus;2) multiple independent loci.

I Theory: If SNP =⇒ X =⇒ Y , then ρYg = ρXgρYX .ρYg

ρXg= ρYX := K , |K | < 1,

independent of g .Similarly, if SNP =⇒ Y =⇒ X , then ...

I Limitation: cannot distinguish X ⇐= SNP =⇒ Y

I Main idea: combining multiple estimates rYg/rXg across g ’s...1) one locus: GLSE;2) multi-loci: IVW ( meta-analysis).

Example: LDL/HDL vs CAD

I Lipid GWAS summary data (Teslovich et al 2010, Nat Genet);CAD GWAS summary data (Schunkert et al 2011, Nat Genet);Reference panel: 489 individuals of EA in the 1000 GenomesProject.

I Partition the genome into 1703 (approximately) independentloci (Berisa and Pickrell 2016, Bioinformatics).

I Consider 8 (or 4) indep loci significant for both LDL (or HDL)and CAD (at p < 5E-6).

I In each locus, pruned out highly correlated SNPs with|r | > 0.8.

LDL vs CAD: Locus 1

[]

●GLS

rs17645031

rs672569

rs599839

rs4970834

0 2 4 6

Mean (95% CI)

SN

P &

GLS

LDL/CAD 5e−06

[]

●GLS

rs17645031

rs672569

rs599839

rs4970834

−1.0 −0.5 0.0 0.5 1.0

Mean (95% CI)

SN

P &

GLS

CAD/LDL 5e−06

Figure: LDL vs CAD locus 1.

LDL vs CAD: Locus 6

[]

●GLS

rs12580178

rs17630235

rs11066028

rs12369009

rs648997

rs7302763

−2 −1 0 1

Mean (95% CI)

SN

P &

GLS

LDL/CAD 5e−06

[]

●GLS

rs12580178

rs17630235

rs11066028

rs12369009

rs648997

rs7302763

−1 0 1

Mean (95% CI)

SN

P &

GLS

CAD/LDL 5e−06

Figure: LDL vs CAD locus 6.

LDL vs CAD: all 8 loci

[]

●IVW

locus8

locus7

locus6

locus5

locus4

locus3

locus2

locus1

0 2 4

Mean (95% CI)

Locu

s &

IVW

LDL/CAD 5e−06

[]

●IVW

locus8

locus7

locus6

locus5

locus4

locus3

locus2

locus1

−1.0 −0.5 0.0 0.5 1.0

Mean (95% CI)

Locu

s &

IVW

CAD/LDL 5e−06

Figure: LDL vs CAD in all 8 loci.

LDL vs CAD: 7 loci

[]

●IVW

locus8

locus7

locus5

locus4

locus3

locus2

locus1

−1 0 1 2 3 4

Mean (95% CI)

Locu

s &

IVW

LDL/CAD 5e−06

[]

●IVW

locus8

locus7

locus5

locus4

locus3

locus2

locus1

−1.0 −0.5 0.0 0.5 1.0

Mean (95% CI)

Locu

s &

IVW

CAD/LDL 5e−06

Figure: LDL vs CAD in 7 loci.

HDL vs CAD: all loci

[]

●IVW

locus4

locus3

locus2

locus1

−2 −1 0 1

Mean (95% CI)

Locu

s &

IVW

HDL/CAD, 5e−6

[]

●IVW

locus4

locus3

locus2

locus1

−1.5 −1.0 −0.5 0.0 0.5 1.0

Mean (95% CI)

Locu

s &

IVW

CAD/HDL, 5e−6

Figure: LDL vs CAD in all 8 loci.

Estimating causal effects

I MR: using SNPs as IVs. IV assumptions:

SNP X Y

×I With a valid IV: βYX =βYg/βXg .

Wald ratios: βYg/βXg for g = 1, 2, ...IVW (meta-analysis for multiple indep SNPs).

I More generally,βYg = βYX βXg + εg ; IVW, PS (Zhao et al 2018)

βYg = β0 + βYX βXg + εg ; Egger reg

βYg = β0g + βYX βXg + εg , β0g ∼iid N(0, τ2); APS/RAPS

I Alternatively, use (weighted) median or mode of the Waldratios (βYg/βXg ’s).

On-going ...

I More applications:LDL/HDL vs CAD: larger datasets;brain imaging ROIs =⇒ AD?

I Co-localization testing

I Fine mapping

I TWAS/2SLS and MR: accounting for SNPs as invalid IVs

I Rare variants (RVs)

I ......

I http://www.biostat.umn.edu/∼weipCode: http://www.biostat.umn.edu/∼weip/prog.htmlR packages aSPU, highmean, GLMaSPU, GEEaSPU,POMaSPU, MiSPU; TLPglm; ...; all on CRAN.Websites with example code:www.wuchong.org/IWAS.htmlwww.wuchong.org/TWAS.html

I This research was supported by NIH, NSF and MSI.

I Many collaborators and (former and current) students!

Thank you!