Analysis Reproduce

Generate the simulation data
- Generate the individual-level data
- Convert the individual-level data into the summary statistics
Run GIFT
- Using individual-level data as input
- Using summary statistics as input
Run FOCUS
Run FOGS
Run MV-IWAS
Run Marginal GWAS
Run Marginal TWAS
- Using individual-level data as input
- Using summary statistics as input

Generate the simulation data #

We randomly selected 50 regions from 1,533 regions. The 1,533 region information is here. The full information about the region can be found in here. For each region in turn, we performed 20 simulation replicates, resulting in a total of 1,000 simulation replicates per setting. For each region, we conducted the simulations based on the realistic genotypes from GEUVADIS (n1=465) and UK Biobank (n2=5,000). Take a region on chr 4 for example. This region includes seven genes: C4orf29, C4orf33, LARP1B, PGRMC2, PHF17, RP11-420A23.1, and SCLT1. We set C4orf29 as the causal gene with the effect size being sqrt(0.1).

Generate the individual-level data #

library("BEDMatrix")
library(mvtnorm)
library(GIFT)

##load the genotypes from GEUVADIS

###load the directory containing the files to be processed only (e.g., plink binary format)
dir <- getwd()
filelocation <- paste0(dir,"/reproduce/simulation_data_generate/GEUVADIS")

###load the directory of plink exe file
plinkexe <- "plink"

###pre-process the file to be a list including gene names vector, cis-genotype matrix and pindex
convert <- pre_process_individual(filelocation, plinkexe)
gene <- convert$gene
Zx <- convert$Z
pindex <- convert$pindex

###please see https://yuanzhongshang.github.io/GIFT/documentation/04_GIFT_Example.html for the details of pre-poccess of genotype data.  

##load the genotypes from UK Biobank
filelocation <- paste0(dir,"/reproduce/simulation_data_generate/UKB")

###pre-process the file to be a list including gene names vector, cis-genotype matrix and pindex
convert <- pre_process_individual(filelocation, plinkexe)
Zy <- convert$Z

##set the correlation of the residual of gene expressions

###We can set the correlation based on the real data or some specific correlation structure 
###Option 1: the correlation below is realistic from GEUVADIS
R <- as.matrix(read.table(paste0(dir,"/reproduce/simulation_data_generate/R.txt")))

###Option 2: generate the correlation from an exponential covariance structure,e.g.ρ=0.9
AR = function(rho, p) {
  outer(1:p, 1:p, FUN = function(x, y) rho^(abs(x - y))) 
}
R <- AR(0.9, length(gene))

##generate the gene expression data 

###set the heritability of gene (PVEzx),here we fixed heritability to be 0.1 for each gene.
PVEzx <- rep(0.1, length(gene))

###generate the effect size from the normal distribution
b=list()
for(i in 1:length(gene)){
  set.seed(i)
  b[[i]] <- matrix(rnorm(pindex[i], 0, sd = sqrt(PVEzx[i]/pindex[i])), pindex[i], 1)
}

###generate the residuals from the multi-normal distribution
set.seed(123)
E <- rmvnorm(dim(Zx)[1], mean = rep(0, length(gene)), sigma = (sqrt(1-PVEzx) %*% t(sqrt(1-PVEzx)))*R)

###sum
pindexsum <- c(0,cumsum(pindex))
X=matrix(0,dim(Zx)[1],length(gene))
for(i in 1:length(gene)){
  X[,i] <- Zx[,(pindexsum[i]+1):pindexsum[i+1]] %*% b[[i]] + E[,i]
}

###set the heritability (PVEzy),here we set RASA1 as the causal gene with the effect size being sqrt(0.1).
PVEzy <- c(0.01,0,0,0,0,0,0)
casual_effect <- sqrt(PVEzy/PVEzx)

###generate the phenotype
set.seed(456)
if (sum(PVEzy) == 0){
  Y <- rnorm(dim(Zy)[2], 0, sqrt(1)) 
} else {
  lp_z=0
  for(i in 1:length(gene)){
    lp_z <- lp_z + casual_effect[i]* as.matrix(Zy[,(pindexsum[i]+1):pindexsum[i+1]]) %*% b[[i]]
  }
  Y <- lp_z + rnorm(dim(Zy)[1], 0, sqrt(1 - sum(PVEzy)))
}

X=scale(X)
Y=scale(Y)
Zx=scale(Zx)
Zy=scale(Zy)

###save these variables
setwd(dir)
save(X, Y, Zx, Zy, gene, pindex, file = "./reproduce/simulation_data_generate/data_generate_individual.RData")

Convert the individual-level data into the summary statistics #

###calculate the z-score from GEUVADIS
n1=dim(Zx)[1]
Zscore1 =NULL
for(i in 1:length(gene)){
  Zscore1 <- cbind(Zscore1,(1/sqrt(n1-1))*t(Zx) %*% X[,i])
}

###calculate the LD matrix from GEUVADIS
LDmatrix1 <- (1/(n1-1))*t(Zx) %*% Zx

###calculate the z-score from UK Biobank
n2=dim(Zy)[1]

###calculate the LD matrix from UK Biobank
LDmatrix2 <- (1/(n2-1))*t(Zy) %*% Zy
Zscore2 <- (1/sqrt(n2-1))* (t(Zy) %*% Y)

###save these variables
save(Zscore1, LDmatrix1, Zscore2, LDmatrix2, R, n1, n2, gene, pindex, file = "./reproduce/simulation_data_generate/data_generate_summary.RData")

Run GIFT #

library(GIFT)

Using individual-level data as input #

##load the simulation data or the real data with the specific format from individual-level data.
###Here we used the simulation data above.
load("./reproduce/simulation_data_generate/data_generate_individual.RData")

##run GIFT
result <- GIFT_individual(X, Y, Zx, Zy, gene, pindex, maxiter=100, tol=1e-3, pleio=0, ncores=1, filter=T)

result
           gene causal_effect            p
1       C4orf29  0.3960856907 0.0001294718
2       C4orf33 -0.0274764328 0.7029436847
3        LARP1B -0.1485976992 0.2228474081
4        PGRMC2  0.1491667252 0.0690383561
5         PHF17 -0.0575508216 0.1795502163
6 RP11-420A23.1  0.0749542623 0.3895547946
7         SCLT1 -0.0003616375 1.0000000000

Using summary statistics as input #

##load the simulation data or the real data with the specific format from summary statistics.
###Here we used the simulation data above.
load("./reproduce/simulation_data_generate/data_generate_summary.RData")

##run GIFT
result <- GIFT_summary(Zscore1, Zscore2, LDmatrix1, LDmatrix2, n1, n2, gene, pindex, R=R, maxiter=100, tol=1e-3, pleio=0, ncores=1, in_sample_LD=T, filter=T, split=5)

result
           gene causal_effect            p
1       C4orf29   0.410449087 9.015665e-06
2       C4orf33   0.007738882 7.971784e-01
3        LARP1B  -0.108524174 2.665893e-01
4        PGRMC2   0.135714320 4.555127e-02
5         PHF17  -0.046589484 1.986309e-01
6 RP11-420A23.1   0.041792289 4.329285e-01
7         SCLT1  -0.013644859 7.362300e-01

Run FOCUS #

FOCUS takes GWAS summary statistics, reference LD, eQTL weight database as inputs. Here we used the same data as above. If you do not have the in-sample LD, you may use the LD from 1000 Genomes reference: https://alkesgroup.broadinstitute.org/FUSION/LDREF.tar.bz2.

library("data.table")
library("gtools")
library(RSQLite)

##load the simulation data or the real data with the specific format from individual-level data.
###Here we used the simulation data.
dir=getwd()
load("./reproduce/simulation_data_generate/data_generate_individual.RData")

##If you need compute own weights, the following additional steps are required.

###integrate the gene expression into the .fam file 
setwd("./reproduce/FOCUS")
system("mkdir weight")
setwd(gsub("/reproduce/FOCUS", "", getwd()))
###FUSION.weights.R is from TWAS/Fusion website:http://gusevlab.org/projects/fusion/
for(i in 1:length(gene)){
  fam <- read.table(paste0("./reproduce/simulation_data_generate/GEUVADIS/",gene[i],".fam"))
  fam <- cbind(fam[,1:5],X[,i])
  write.table(fam,paste0("./reproduce/simulation_data_generate/GEUVADIS/",gene[i],".fam"), quote=F,sep=" ",col.names=F,row.names=F)
  system(paste0("Rscript FUSION.weights.R --bfile ./reproduce/simulation_data_generate/GEUVADIS/",gene[i]," --tmp ./reproduce/FOCUS/tmp --out ./reproduce/FOCUS/weight/",gene[i]," --verbose 0 --models bslmm,lasso,top1,enet,blup"))
}

##the function outputs above and the downloaded existing weight often in the .wgt.RDat format (for example, [http://gusevlab.org/projects/fusion/](http://gusevlab.org/projects/fusion/))

###conduct .db

####model.txt
setwd("./reproduce/FOCUS/weight")
ot <- mixedsort(list.files(".",full.names=F))
n <- length(ot)
model <- cbind.data.frame(c(1:n), rep("bslmm",n), rep(1,n), c(1:n))
colnames(model) <- c("id", "inference", "ref_id", "mol_id")
setwd(gsub("/weight", "", getwd()))
system("mkdir db")
setwd("./db")
write.table(model, "model.txt", row.names = F,quote = F, col.names = T, sep = "\t")

####weight.txt
wt = att = list()
setwd(gsub("/db", "/weight", getwd()))
for(i in 1:n){
    onfi <- ot[i]
    load(onfi)
    genefile <- gsub(".wgt.RDat","",onfi)
    bim <- as.data.frame(fread(paste0(dir,"/reproduce/simulation_data_generate/GEUVADIS/",genefile,".bim"),header=F))
    p <- nrow(bim)
    ##Here we choose the BSLMM weight.
    all <- cbind.data.frame(bim[,2],bim[,1],bim[,4:6],wgt.matrix[,colnames(wgt.matrix)=="bslmm"],rep(NA,p),rep(i,p))
    colnames(all) <- c("snp","chrom","pos","effect_allele","alt_allele","effect","std_error","model_id")
    wt[[i]] <- all
    cv <- cbind.data.frame(c("cv.R2","cv.R2.pval"),cv.performance[,1],rep(i,2))
    colnames(cv) <- c("attr_name","value","model_id")
    att[[i]] <- cv
  }
rb <- Reduce(rbind,wt)
N <- nrow(rb)
All <- cbind.data.frame(c(1:N),rb)
colnames(All)[1] <- "id"
setwd(gsub("/weight", "/db", getwd()))
write.table(All, "weight.txt", row.names = F,quote = F, col.names = T, sep = "\t")

####modelattribute.txt
modelatt <- Reduce(rbind, att)
N <- nrow(modelatt)
att.txt <- cbind.data.frame(c(1:N), modelatt)
colnames(att.txt)[1] <- "id"
write.table(att.txt, "modelattribute.txt", row.names = F, quote = F, col.names = T,sep = "\t")

####molecularfeature.txt
o1 <- gsub(".wgt.RDat","",ot)
infos=as.data.frame(fread(paste0(dir,"/reproduce/simulation_data_generate/geneinfo.txt"),header=T))

###We focused on protein coding genes and lincRNAs that are annotated in GENCODE (release 12).
###please see https://github.com/yuanzhongshang/GIFT/blob/main/reproduce/gencodev12.tsv for the information of 15,577 genes included in this research.
info <- infos[infos$genetype2 %in% o1,]
sub <- info[order(match(info$genetype2,o1)),]
mole <- cbind.data.frame(c(1:n), sub[,5], rep("<NA>",n), sub[,4], sub[,3], sub[,6], sub[,1:2])
colnames(mole) <- c("id", "ens_gene_id", "ens_tx_id", "mol_name", "type", "chrom", "tx_start", "tx_stop")
write.table(mole, "molecularfeature.txt",row.names = F, quote = F, col.names = T, sep = "\t")

####refpanel.txt
myref <- cbind.data.frame(1, "GEUVADIS", "blood", "rnaseq")
colnames(myref) <- c("id", "ref_name", "tissue", "assay")
write.table(myref, "refpanel.txt", row.names = F, quote = F, col.names = T,sep = "\t")
  
####construct FOCUS.db file
model <- as.data.frame(fread("model.txt",header = T))
modelattribute <- as.data.frame(fread("modelattribute.txt", header = T))
molecularfeature <- as.data.frame(fread("molecularfeature.txt", header = T))
refpanel <- as.data.frame(fread("refpanel.txt",header = T))
weight <- as.data.frame(fread("weight.txt",header = T))

setwd(gsub("/db", "", getwd()))
mydb <- dbConnect(RSQLite::SQLite(), "weight.db")
dbWriteTable(mydb, "model", model)
dbWriteTable(mydb, "modelattribute", modelattribute)
dbWriteTable(mydb, "molecularfeature",molecularfeature)
dbWriteTable(mydb, "refpanel", refpanel)
dbWriteTable(mydb, "weight", weight)

##perform GWAS using PLINK or process the obtained GWAS summary statistics
##If you perform GWAS using other software, please provide a .txt file contains columns: "CHR", "SNP", "BP", "A1", "A2", "Z", "P", "N".
##conduct the pheotype file used in plink
FID <- read.table(paste0(dir,"/reproduce/simulation_data_generate/Zy.fam"))$V1
IID <- FID
data <- cbind(FID,IID,Y)
colnames(data)[3] <- "V1"
write.table(data,"gwasY.txt",quote=F,row.names=F)

setwd(dir)
system("plink --bfile ./reproduce/simulation_data_generate/Zy --pheno  ./reproduce/FOCUS/gwasY.txt --pheno-name V1  --allow-no-sex --assoc --out ./reproduce/FOCUS/gwas")
data <- fread("./reproduce/FOCUS/gwas.qassoc")
bim <- fread("./reproduce/simulation_data_generate/Zy.bim")
summary <- cbind(data[,1:3], bim[,5:6], data[,8], data[,9], data[,4])
colnames(summary) <- c("CHR", "SNP", "BP", "A1", "A2", "Z", "P", "N")
write.table(summary,"./reproduce/FOCUS/gwas.txt",quote=F,row.names=F)

###clean GWAS summary data
system("focus munge ./reproduce/FOCUS/gwas.txt --output ./reproduce/FOCUS/GWAS.cleaned")

##run FOCUS
system("focus finemap ./reproduce/FOCUS/GWAS.cleaned.sumstats.gz ./reproduce/simulation_data_generate/Zy ./reproduce/FOCUS/weight.db --locations 37:EUR --chr 4 --start 128996665 --stop 130591885 --prior-prob ./reproduce/gencodev12.tsv --p-threshold 1 --out ./reproduce/FOCUS/result")

result <- read.table("./reproduce/FOCUS/result.focus.tsv", header = T)

result
block	ens_gene_id	ens_tx_id	mol_name	tissue	ref_name	type	chrom	tx_start	tx_stop	block_genes	trait	inference_pop1	inter_z_pop1	cv.R2_pop1	cv.R2.pval_pop1	ldregion_pop1	twas_z_pop1	pips_pop1	in_cred_set_pop1
4:128996665-4:130591885	ENSG00000164074.8	<NA>	C4orf29	blood	GEUVADIS	protein_coding	4	128786435	129060866	7	trait	bslmm	NA	0.0318458223286902	6.44829004061019e-05	4:129000314-4:130133779	4.77	0.998	1
4:128996665-4:130591885	ENSG00000077684.10	<NA>	PHF17	blood	GEUVADIS	protein_coding	4	129630779	129896379	7	trait	bslmm	NA	0.191771884804571	1.99043122525914e-23	4:129000314-4:130133779	-1.66	0.123	1
4:128996665-4:130591885	ENSG00000138709.13	<NA>	LARP1B	blood	GEUVADIS	protein_coding	4	128882423	129244086	7	trait	bslmm	NA	0.0551955272133811	1.78083420363514e-07	4:129000314-4:130133779	-3.1	0.115	1
4:128996665-4:130591885	ENSG00000151466.7	<NA>	SCLT1	blood	GEUVADIS	protein_coding	4	129686076	130114764	7	trait	bslmm	NA	0.0959445604427946	5.09993685259739e-12	4:129000314-4:130133779	-1.11	0.0683	1
4:128996665-4:130591885	ENSG00000164040.10	<NA>	PGRMC2	blood	GEUVADIS	protein_coding	4	129090392	129309984	7	trait	bslmm	NA	0.0742207822860967	1.40211140963927e-09	4:129000314-4:130133779	-1.96	0.051	0
4:128996665-4:130591885	ENSG00000251432.2	<NA>	RP11-420A23.1	blood	GEUVADIS	lincRNA	4	129113906	129540549	7	trait	bslmm	NA	-0.00166001190601528	0.630987463676111	4:129000314-4:130133779	-1.81	0.0486	0
4:128996665-4:130591885	ENSG00000151470.7	<NA>	C4orf33	blood	GEUVADIS	protein_coding	4	129914472	130134487	7	trait	bslmm	NA	0.0123982415772871	0.00928127102852506	4:129000314-4:130133779	0.106	0.0408	0
4:128996665-4:130591885	NULL.MODEL	NA	NULL	NA	NA	NULL	4	NA	NA	7	trait	NA	NA	NA	NA	4:129000314-4:130133779	0	0.000349	0

Run FOGS #

FOGS takes eQTL-derived weights, reference LD, GWAS summary statistics and gene list as inputs.

library("data.table")
library(RSQLite)
library(gtools)

##load the pindex from the simulation data.
##You can also load from the weight file directly.
dir=getwd()
load("./reproduce/simulation_data_generate/data_generate_individual.RData")

##prepare the eQTL-derived weights
weights <- "./reproduce/FOCUS/weight.db"
weightsave <- paste0(dir,"/reproduce/FOGS/weight.txt")
sqlite.driver <- dbDriver("SQLite")
db <- dbConnect(sqlite.driver,dbname = weights)
dbListTables(db)
weights <- dbReadTable(db, "weight")

setwd("./reproduce/FOCUS/weight")
ot <- mixedsort(list.files(".",full.names=F))
dn <- NULL
for(i in 1:length(gene)){
  onfi <- ot[i]
  gen1 <- gsub(".wgt.RDat","",onfi)
  dn <- c(dn, rep(gen1, pindex[which(gene %in% gen1)]))
}
weightsn <- cbind(weights, dn)
weightsn <- weightsn[,c(2,10,7,6,5)]
colnames(weightsn) <- c("rsid", "gene", "weight", "ref_allele", "eff_allele")
write.table(weightsn, weightsave, col.names = T, row.names = F, quote = F)

weights <- paste0(dir,"/reproduce/FOGS/weight.txt")

##prepare the summary statistics
##If you perform GWAS using other software, please provide a .txt file contains columns: "CHR", "SNP", "POS", "A1", "A2", "beta", "se", "Z", "P", "samplesize".
setwd(paste0(dir, "/reproduce/FOGS"))
source("munge.R")

data <- fread(paste0(dir,"/reproduce/FOCUS/gwas.qassoc"))
bim <- fread(paste0(dir,"/reproduce/simulation_data_generate/Zy.bim"))
summary <- cbind(data[,1:3], bim[,5:6], data[,5:6], data[,8], data[,9],data[,4])
colnames(summary) <- c("CHR", "SNP", "POS", "A1", "A2", "beta", "se", "Z", "P", "samplesize")
write.table(summary, "gwas.txt", quote = F, row.names = F)

sumstats <- paste0(dir,"/reproduce/FOGS/gwas.txt")
data_processed <- munge_sumstat(sumstats)
write.table(data_processed, "gwas_processed_data.txt", col.names = T, row.names = F, quote = F)

sumstat <- paste0(dir,"/reproduce/FOGS/gwas_processed_data.txt")

##load the directory of LD reference
##Here we used the same data as above.
refld <- paste0(dir,"/reproduce/simulation_data_generate/Zy")

##load the directory of gene list
genelist <- paste0(dir,"/reproduce/FOGS/genelist15577.txt")

##match the locus in LDetect region used in the FOGS
ldetect <- read.table(paste0(dir,"/reproduce/FOGS/LDetectregion.txt"),header=T)
chr_id <- 4
locus_id <- which(ldetect[ldetect$chr=="chr4",]$start==128996665)

##other inputs
outd <- paste0(dir,"/reproduce/FOGS/result")
saveprefix <- paste0(dir,"/reproduce/FOGS/result")
loci <-paste0(dir,"/reproduce/LDetect/EUR/")

##run FOGS
system(paste0("Rscript FOGS.R --refld ",refld," --outd ",outd," --loci ",loci," --weights ",weights," --genelist ",genelist," --sumstat ",sumstat," --saveprefix ",saveprefix," --chr_id ",chr_id," --locus_id ",locus_id))

result <- read.table(paste0(dir,"/reproduce/FOGS/resultCHR_4_Locus",locus_id,".txt"),header = T)

result
CHR ID P0 P1 n.SNP n.condSNP FOGS-aSPU TWAS Focus Runtime(s)
4 C4orf33 129914472 130134487 239 112 0.948051948051948 0.912378129020417 0.00015761489003644 24.856
4 RP11-420A23.1 129113906 129540549 252 91 0.68031968031968 0.0666979609501761 0.000805957175765842 18.123
4 SCLT1 129686076 130114764 324 101 0.405594405594406 0.261608296740732 0.000289547991162607 27.113
4 PHF17 129630779 129896379 163 120 0.175824175824176 0.0951010742555474 0.00060903461396202 17.343
4 PGRMC2 129090392 129309984 103 131 0.010989010989011 0.0493022814419665 0.00102967099297041 12.53
##Of note, FOGS fails to perform one gene when the cis-SNPs of one gene are fully contained in the cis-SNPs of another gene. 
##Here cis-SNPs of C4orf29 and LARP1B are included in that of C4orf33. Thus, the result does not contain these two genes.

Run MV-IWAS #

MV-IWAS takes GWAS summary statistics, reference LD, eQTL weight database as inputs. Here we used the same data as above.

library(MVIWAS)

##load the simulation data or the real data with the specific format from the summary statistics
###Here we used the simulation data.
dir=getwd()
load("./reproduce/simulation_data_generate/data_generate_summary.RData")

##prepare the eQTL-derived weights
p=sum(pindex)

##Here we used the BSLMM weight for the fair comparison.
betax <- NULL
setwd("./reproduce/FOCUS/weight")
for(i in 1:length(gene)){
  if(i == length(gene)){
    load(paste0(gene[i],".wgt.RDat"))
    betax1 <- wgt.matrix[,colnames(wgt.matrix)=="bslmm"]
    betax <- c(betax,betax1)
  }else{
   load(paste0(gene[i],".wgt.RDat"))
   betax1 <- wgt.matrix[,colnames(wgt.matrix)=="bslmm"]
   betax <- c(betax, betax1, rep(0,p))
  }
}
betax <- matrix(betax, p, length(gene))

betay <- Zscore2/sqrt(n2-1)
se_betaZY <- matrix(1/sqrt(n2-1),p,1)

##run MV-IWAS
result <- mv_iwas_summ(betay, se_betaZY, betax, LDmatrix2, n2, "Continuous", n_case = NULL, n_control = NULL)
result$TERM <- gene
write.table(result,paste0(dir,"/reproduce/MVIWAS/result.txt"), quote = F, row.names = F)

result
    MODEL          TERM         BETA         SE          Z            P NSNP
1 MV-IWAS       C4orf29  0.221597362 0.05474690  4.0476699 5.173003e-05 1231
2 MV-IWAS       C4orf33 -0.008441355 0.06985365 -0.1208434 9.038150e-01 1231
3 MV-IWAS        LARP1B -0.105409347 0.06662337 -1.5821678 1.136113e-01 1231
4 MV-IWAS        PGRMC2  0.055406960 0.05005174  1.1069938 2.682966e-01 1231
5 MV-IWAS         PHF17 -0.024894469 0.02305678 -1.0797030 2.802745e-01 1231
6 MV-IWAS RP11-420A23.1  0.062661326 0.12442069  0.5036246 6.145252e-01 1231
7 MV-IWAS         SCLT1 -0.005552580 0.03470806 -0.1599796 8.728972e-01 1231
  DiseaseType
1  Continuous
2  Continuous
3  Continuous
4  Continuous
5  Continuous
6  Continuous
7  Continuous

##Note that, matrix objects now also inherit from class "array" since R 4.0.0. mv_iwas_summ() contains some class checks which should modify to avoid the bug.
##If you used the R verion over 4.0.0, source("./reproduce/MVIWAS/mv_iwas_summ.R")

Run Marginal GWAS #

We used PLINK-1.9 to perform the GWAS analysis.

library("data.table")

##conduct the pheotype file used in plink
dir <- getwd()
FID <- read.table(paste0(dir,"/reproduce/simulation_data_generate/Zy.fam"))$V1
IID <- FID
data <- cbind(FID,IID,Y)
colnames(data)[3] <- "V1"
write.table(data,"gwasY.txt",quote=F,row.names=F)

setwd(dir)
system("plink --bfile ./reproduce/simulation_data_generate/Zy --pheno  ./reproduce/FOCUS/gwasY.txt --pheno-name V1  --allow-no-sex --assoc --out ./reproduce/FOCUS/gwas")
data <- fread("./reproduce/FOCUS/gwas.qassoc")

head(data)
   CHR         SNP        BP NMISS     BETA      SE        R2      T         P
1:   4  rs34350456 129000314  5000 -0.07705 0.02115 0.0026490 -3.644 0.0002717
2:   4   rs1993722 129002172  5000 -0.07705 0.02115 0.0026490 -3.644 0.0002717
3:   4   rs7684955 129007262  5000 -0.07705 0.02115 0.0026490 -3.644 0.0002717
4:   4 rs116213612 129011061  5000  0.06143 0.03414 0.0006473  1.799 0.0720400
5:   4   rs2306054 129012181  5000 -0.07705 0.02115 0.0026490 -3.644 0.0002717
6:   4   rs1064205 129012638  5000 -0.07705 0.02115 0.0026490 -3.644 0.0002717

Run Marginal TWAS #

Using individual-level data as input #

library(gtools)
library("BEDMatrix")

##Here we also used the BSLMM weight above.
dir=getwd()
setwd("./reproduce/FOCUS/weight")

ot <- mixedsort(list.files(".",full.names=F))
gene <- NULL
for(i in 1:length(ot)){
  onfi <- ot[i]
  gene[i] <- gsub(".wgt.RDat","",onfi)
}

##impute function
##We impute the genotypes from the UK Biobank.
impu <- function(x) {
  x[is.na(x)] <- mean(x, na.rm = TRUE)
  x
}

##load the phenotype
setwd(dir)
load("./reproduce/simulation_data_generate/data_generate_individual.RData")

Z <- NULL
P <- NULL
setwd("./reproduce/FOCUS/weight")
for(i in 1:length(gene)){
  load(ot[i])
  beta <- wgt.matrix[,colnames(wgt.matrix)=="bslmm"]
  Zy <- BEDMatrix(paste0(dir,"/reproduce/simulation_data_generate/UKB/",gene[i]))
  Zy <- apply(Zy ,2, impu)
  p <- dim(Zy)[2]
  tmp <- cbind(Zy,Y)
  tmp <- scale(na.omit(tmp))
  Zy <- tmp[,1:p]
  Y <- tmp[,p+1]
  pred <- Zy %*% beta
  fit <- lm(Y ~ pred)
  Z[i] <- coefficients(summary(fit))[2,3]
  P[i] <- coefficients(summary(fit))[2,4]
}
result <- data.frame(gene, Z, P)
write.table(result,paste0(dir, "/reproduce/TWAS/result_individual.txt"), quote = F, row.names = F)

result
           gene          Z            P
1       C4orf29  4.9493385 7.691249e-07
2       C4orf33  0.1629571 8.705588e-01
3        LARP1B -3.1384610 1.708262e-03
4        PGRMC2 -1.9652850 4.943655e-02
5         PHF17 -1.6692476 9.513096e-02
6 RP11-420A23.1 -1.8330907 6.684853e-02
7         SCLT1 -1.1222759 2.617990e-01

Using summary statistics as input #

##load the simulation data or the real data with the specific format from the summary statistics
###Here we used the simulation data.
dir=getwd()
load("./reproduce/simulation_data_generate/data_generate_summary.RData")

Z <- NULL
P <- NULL
pindexsum <- c(0,cumsum(pindex))
setwd("./reproduce/FOCUS/weight")
for(i in 1:length(gene)){
  load(ot[i])
  beta <- wgt.matrix[,colnames(wgt.matrix)=="bslmm"]
  Z[i] <- sum(beta * Zscore2[(pindexsum[i]+1):pindexsum[i+1]])/sqrt(sum(beta*(LDmatrix2[(pindexsum[i]+1):pindexsum[i+1],(pindexsum[i]+1):pindexsum[i+1]] %*% beta)))
  P[i] <- 2*pnorm(-abs(Z[i]), 0, 1)
}
result <- data.frame(gene, Z, P)
write.table(result,paste0(dir, "/reproduce/TWAS/result_summary.txt"), quote = F, row.names = F)

result
           gene         Z            P
1       C4orf29  4.937748 7.902983e-07
2       C4orf33  0.162973 8.705397e-01
3        LARP1B -3.135687 1.714523e-03
4        PGRMC2 -1.964723 4.944635e-02
5         PHF17 -1.668949 9.512741e-02
6 RP11-420A23.1 -1.832658 6.685343e-02
7         SCLT1 -1.122247 2.617575e-01

Analysis Reproduce

Table of Contents

Generate the simulation data #

Generate the individual-level data #

Convert the individual-level data into the summary statistics #

Run GIFT #

Using individual-level data as input #

Using summary statistics as input #

Run FOCUS #

Run FOGS #

Run MV-IWAS #

Run Marginal GWAS #

Run Marginal TWAS #

Using individual-level data as input #

Using summary statistics as input #