functions.R 30.7 KB
Newer Older
Rauschenberger's avatar
Rauschenberger committed
1

Rauschenberger's avatar
Rauschenberger committed
2
3
4
5
6
7
8
9
10
11
#' @export
#' @title
#' Get SNP data (Geuvadis)
#' 
#' @description
#' This function transforms SNP data (local machine).
#' 
#' @param chr
#' chromosome: integer \eqn{1-22}
#' 
Rauschenberger's avatar
Rauschenberger committed
12
#' @param data
Rauschenberger's avatar
Rauschenberger committed
13
14
#' local directory for VCF (variant call format) and SDRF (sample and data relationship format) files
#' 
Rauschenberger's avatar
Rauschenberger committed
15
16
17
#' @param path
#' local directory for output
#' 
Rauschenberger's avatar
Rauschenberger committed
18
19
20
#' @examples
#' path <- "C:/Users/a.rauschenbe/Desktop/spliceQTL/data"
#' 
Rauschenberger's avatar
Rauschenberger committed
21
get.snps.geuvadis <- function(chr,data=NULL,path=getwd()){
Rauschenberger's avatar
Rauschenberger committed
22
    
Rauschenberger's avatar
Rauschenberger committed
23
24
25
26
27
28
29
30
31
32
33
34
    if(is.null(data)){
        data <- path
        # download SNP data
        file <- paste0("GEUVADIS.chr",chr,".PH1PH2_465.IMPFRQFILT_BIALLELIC_PH.annotv2.genotypes.vcf.gz")
        url <- paste0("http://www.ebi.ac.uk/arrayexpress/files/E-GEUV-1/genotypes/",file)
        destfile <- file.path(data,file)
        if(!file.exists(destfile)){
            utils::download.file(url=url,destfile=destfile,method="auto")
        }
        # transform with PLINK
        setwd(data)
        system(paste0("plink --vcf GEUVADIS.chr",chr,".PH1PH2_465.IMPFRQFILT_BIALLELIC_PH.annotv2.genotypes.vcf.gz",
Rauschenberger's avatar
Rauschenberger committed
35
                  " --maf 0.05 --geno 0 --make-bed --out snps",chr),invisible=FALSE)
Rauschenberger's avatar
Rauschenberger committed
36
37
38
39
40
41
42
43
        # obtain identifiers
        file <- "E-GEUV-1.sdrf.txt"
        url <- paste("http://www.ebi.ac.uk/arrayexpress/files/E-GEUV-1/",file,sep="")
        destfile <- file.path(data,file)
        if(!file.exists(destfile)){
            utils::download.file(url=url,destfile=destfile,method="auto")
        }
    }
Rauschenberger's avatar
Rauschenberger committed
44
45
    
    # read into R
Rauschenberger's avatar
Rauschenberger committed
46
47
48
    bed <- file.path(data,paste("snps",chr,".bed",sep=""))
    bim <- file.path(data,paste("snps",chr,".bim",sep=""))
    fam <- file.path(data,paste("snps",chr,".fam",sep=""))
Rauschenberger's avatar
Rauschenberger committed
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
    X <- snpStats::read.plink(bed=bed,bim=bim,fam=fam)
    X$fam <- NULL; all(diff(X$map$position) > 0)
    
    # fitler MAF
    maf <- snpStats::col.summary(X$genotypes)$MAF
    cond <- maf >= 0.05
    X$genotypes <- X$genotypes[,cond]
    X$map <- X$map[cond,]
    
    # format
    colnames(X$genotypes) <- paste0(X$map$chromosome,":",X$map$position)
    snps <- methods::as(object=X$genotypes,Class="numeric")
    class(snps) <- "integer"
    
    # change identifiers
Rauschenberger's avatar
Rauschenberger committed
64
    samples <- utils::read.delim(file=file.path(data,"E-GEUV-1.sdrf.txt"))
Rauschenberger's avatar
Rauschenberger committed
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
    match <- match(rownames(snps),samples$Source.Name)
    rownames(snps) <- samples$Comment.ENA_RUN.[match]
    snps <- snps[!is.na(rownames(snps)),]
    
    save(object=snps,file=file.path(path,paste0("Geuvadis.chr",chr,".RData")))
}


#' @export
#' @title
#' Get SNP data (BBMRI)
#' 
#' @description
#' This function transforms SNP data (virtual machine).
#' 
#' @param chr
#' chromosome: integer \eqn{1-22}
#' 
#' @param biobank
#' character "CODAM", "LL", "LLS", "NTR", "PAN", "RS", or NULL (all)
#' 
#' @param path
#' data directory
#' 
Rauschenberger's avatar
Rauschenberger committed
89
#' @param size
Rauschenberger's avatar
Rauschenberger committed
90
91
#' maximum number of SNPs to read in at once;
#' trade-off between memory usage (low) and speed (high)
Rauschenberger's avatar
Rauschenberger committed
92
#' 
Rauschenberger's avatar
Rauschenberger committed
93
94
95
#' @examples
#' path <- "/virdir/Scratch/arauschenberger/trial"
#'
Rauschenberger's avatar
Rauschenberger committed
96
get.snps.bbmri <- function(chr,biobank=NULL,path=getwd(),size=500*10^3){
Rauschenberger's avatar
Rauschenberger committed
97
98
99
100
101
102
103
104
105
106
107
108

    start <- Sys.time()
    message(rep("-",times=20)," chromosome ",chr," ",rep("-",times=20))
    
    p <- 5*10^6 # (maximum number of SNPs per chromosome, before filtering)
    skip <- seq(from=0,to=p,by=size)
    if(is.null(biobank)){
        study <- c("CODAM","LL","LLS0","LLS1","NTR0","NTR1","PAN","RS")
    } else if(biobank=="LLS"){
        study <- c("LLS0","LLS1")
    } else if(biobank=="NTR"){
        study <- c("NTR0","NTR1")
Rauschenberger's avatar
Rauschenberger committed
109
    } else if(biobank %in% c("CODAM","LL","PAN","RS")){
Rauschenberger's avatar
Rauschenberger committed
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
        study <- biobank
    } else{
        stop("Invalid biobank.",call.=FALSE)
    }
    collect <- matrix(list(),nrow=length(skip),ncol=length(study))
    colnames(collect) <- study
    
    for(i in seq_along(skip)){
        message("\n","chunk ",i,": ",appendLF=FALSE)
        for(j in seq_along(study)){
            message(study[j],"  ",appendLF=FALSE)
            
            # Locating files on virtual machine.
            dir <- study[j]
            if(study[j]=="LLS0"){dir <- "LLS/660Q"}
            if(study[j]=="LLS1"){dir <- "LLS/OmniExpr"}
            if(study[j]=="NTR0"){dir <- "NTR/Affy6"}
            if(study[j]=="NTR1"){dir <- "NTR/GoNL"}
            path0 <- file.path("/mnt/virdir/Backup/RP3_data/HRCv1.1_Imputation",dir)
Rauschenberger's avatar
Rauschenberger committed
129
            path1 <- path
Rauschenberger's avatar
Rauschenberger committed
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
            file0 <- paste0("chr",chr,".dose.vcf.gz")
            file1 <- paste0(study[j],".chr",chr,".dose.vcf.gz")
            file2 <- paste0(study[j],".chr",chr,".dose.vcf")
            
            # Decompressing missing files in personal folder.
            if(!file.exists(file.path(path1,file2))){
                file.copy(from=file.path(path0,file0),to=file.path(path1,file1))
                R.utils::gunzip(filename=file.path(path1,file1),remove=TRUE,overwrite=TRUE)
            }
            
            # Reading in files.
            vcf <- vcfR::read.vcfR(file=file.path(path1,file2),skip=skip[i],nrows=size,verbose=FALSE)
            vcf <- vcf[vcf@fix[,"CHROM"]!="",] # bug fix
            vcf@fix[,"ID"] <- paste0(vcf@fix[,"ID"],"_",seq_len(dim(vcf)["variants"]))
            collect[i,j][[1]] <- vcf
            stop <- dim(vcf)["variants"]==0
            final <- dim(vcf)["variants"]<size
            if(stop){break}
        }
        print(utils::object.size(collect),units="Gb")
        end <- Sys.time()
        if(stop){break}
        
        # Calculating minor allele frequency.
        num <- numeric(); maf <- list()
        for(j in seq_along(study)){
            num[j] <- dim(collect[i,j][[1]])["gt_cols"] # replace by adjusted sample sizes?
            maf[[j]] <- num[j]*vcfR::maf(collect[i,j][[1]])[,"Frequency"]
        }
        cond <- rowSums(do.call(what="cbind",args=maf))/sum(num)>0.05
        if(sum(cond)==0){if(final){break}else{next}}
        
        # Filtering out genotypes.
        for(j in seq_along(study)){
            gt <- vcfR::extract.gt(collect[i,j][[1]][cond,])
            gt[gt=="0|0"] <- 0
            gt[gt=="0|1"|gt=="1|0"] <- 1
            gt[gt=="1|1"] <- 2
            storage.mode(gt) <- "integer"
            collect[i,j][[1]] <- gt
        }
        
        if(final){break}
    }
    
    # Removing empty rows.
    cond <- apply(collect,1,function(x) all(sapply(x,length)==0))
    collect <- collect[!cond,,drop=FALSE]
Rauschenberger's avatar
Rauschenberger committed
178
    #save(object=collect,file=file.path(path1,paste0("temp.chr",chr,".RData")))
Rauschenberger's avatar
Rauschenberger committed
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
    #load(file.path(path1,paste0("temp.chr",chr,".RData")))
    
    # Fusing all matrices.
    snps <- NULL
    for(i in seq_len(nrow(collect))){
        inner <- NULL
        for(j in seq_len(ncol(collect))){
            add <- collect[i,j][[1]]
            colnames(add) <- paste0(colnames(collect)[j],":",colnames(add))
            inner <- cbind(inner,add)
        }
        snps <- rbind(snps,inner)
    }
    attributes(snps)$time <- end-start
    rownames(snps) <- sapply(strsplit(x=rownames(snps),split="_"),function(x) x[[1]])
    snps <- t(snps)
    
    # Filter samples.
    rownames(snps) <- sub(x=rownames(snps),pattern="LLS0|LLS1",replacement="LLS")
    rownames(snps) <- sub(x=rownames(snps),pattern="NTR0|NTR1",replacement="NTR")
Rauschenberger's avatar
Rauschenberger committed
199

Rauschenberger's avatar
Rauschenberger committed
200
201
202
203
204
    if(is.null(biobank)){
        save(object=snps,file=file.path(path1,paste0("BBMRI.chr",chr,".RData")))
    } else {
        save(object=snps,file=file.path(path1,paste0(biobank,".chr",chr,".RData")))
    }
Rauschenberger's avatar
Rauschenberger committed
205
206
207
    
    # Remove temporary files.
    for(j in seq_along(study)){
Rauschenberger's avatar
Rauschenberger committed
208
        file.remove(file.path(path1,paste0(study[j],".chr",chr,".dose.vcf")))
Rauschenberger's avatar
Rauschenberger committed
209
210
    }
    
Rauschenberger's avatar
Rauschenberger committed
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
}


#' @export
#' @title
#' Get exon data (Geuvadis)
#' 
#' @description
#' This function transforms exon data (virtual machine).
#' 
#' @param path
#' data directory 
#' 
#' @examples
#' path <- "/virdir/Scratch/arauschenberger/trial"
#' 
get.exons.geuvadis <- function(path=getwd()){

    nrows <- 303544
    file <-"/virdir/Scratch/rmenezes/data_counts.txt"
    exons <- utils::read.table(file=file,header=TRUE,nrows=nrows)
    exons <- exons[exons[,"chr"] %in% 1:22,] # autosomes
    rownames(exons) <- exon_id <- paste0(exons[,"chr"],"_",exons[,"start"],"_",exons[,"end"])
    gene_id <- as.character(exons[,4])
    exons <- t(exons[,-c(1:4)])

    save(list=c("exons","exon_id","gene_id"),file=file.path(path,"Geuvadis.exons.RData"))
}


#' @export
#' @title
#' Get exon data (BBMRI)
#' 
#' @description
#' This function transforms exon data (virtual machine).
#' 
#' @param path
#' data directory 
#' 
#' @examples
#' path <- "/virdir/Scratch/arauschenberger/trial"
#' 
get.exons.bbmri <- function(path=getwd()){
    
    # sample identifiers:
    # (1) loading quality controlled gene expression data 
    # (2) extracting sample identifiers
    # (3) removing identifiers without SNP data
    # (4) translating identifiers
    utils::data(rnaSeqData_ReadCounts_BIOS_cleaned,package="BBMRIomics") # (1)
Rauschenberger's avatar
Rauschenberger committed
262
263
    cd <- SummarizedExperiment::colData(counts)[,c("biobank_id","imputation_id","run_id")] # (2)
    counts <- NULL
Rauschenberger's avatar
Rauschenberger committed
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
    names(cd) <- substr(names(cd),start=1,stop=3) # abbreviate names
    cd <- cd[!is.na(cd$imp),] # (3)
    cd$id <- NA # (4)
    cd$id[cd$bio=="CODAM"] <- sapply(strsplit(x=cd$imp[cd$bio=="CODAM"],split="_"),function(x) x[[2]])
    cd$id[cd$bio=="LL"] <- sub(pattern="1_LLDeep_",replacement="",x=cd$imp[cd$bio=="LL"])
    cd$id[cd$bio=="LLS"] <- sapply(strsplit(x=cd$imp[cd$bio=="LLS"],split="_"),function(x) x[[2]])
    cd$id[cd$bio=="NTR"] <- sapply(strsplit(x=cd$imp[cd$bio=="NTR"],split="_"),function(x) x[[2]])
    cd$id[cd$bio=="PAN"] <- cd$imp[cd$bio=="PAN"]
    cd$id[cd$bio=="RS"] <- sub(pattern="RS1_|RS2_|RS3_",replacement="",x=cd$imp[cd$bio=="RS"])
    
    # Identify individual not with "id" but with "bio:id".
    any(duplicated(cd$id)) # TRUE
    sapply(unique(cd$bio),function(x) any(duplicated(cd$id[x]))) # FALSE
    
    # exon data:
    # (1) loading exon expression data
    # (2) extracting sample identifiers
    # (3) retaining autosomes
    # (4) retaining samples from above
    load("/virdir/Backup/RP3_data/RNASeq/v2.1.3/exon_base/exon_base_counts.RData") # (1)
    colnames(counts) <- sub(pattern=".exon.base.count.gz",replacement="",x=colnames(counts)) # (2)
    autosomes <- sapply(strsplit(x=rownames(counts),split="_"),function(x) x[[1]] %in% 1:22) # (3)
    exons <- counts[autosomes,cd$run] # (3) and (4)
    exon_id <- exon_id[autosomes] # (3)
    gene_id <- gene_id[autosomes] # (3)
    colnames(exons) <- paste0(cd$bio,":",cd$id)
    exons <- t(exons)
    
    save(list=c("exons","exon_id","gene_id"),file=file.path(path,"BBMRI.exons.RData"))
}


Rauschenberger's avatar
Rauschenberger committed
296
297
298
299
300
#' @export
#' @title
#' Prepare data matrices
#' 
#' @description
Rauschenberger's avatar
Rauschenberger committed
301
302
#' This function removes duplicate samples from each matrix,
#' only retains samples appearing in all matrices,
Rauschenberger's avatar
Rauschenberger committed
303
#' and brings samples into the same order.
Rauschenberger's avatar
Rauschenberger committed
304
#' 
Rauschenberger's avatar
Rauschenberger committed
305
#' @param ...
Rauschenberger's avatar
Rauschenberger committed
306
307
#' matrices with samples in the rows and variables in the columns,
#' with sample identifiers as rows names
Rauschenberger's avatar
Rauschenberger committed
308
#' 
Rauschenberger's avatar
Rauschenberger committed
309
310
#' @param message
#' display messages\strong{:} logical
Rauschenberger's avatar
Rauschenberger committed
311
312
#' 
#' @examples
Rauschenberger's avatar
Rauschenberger committed
313
314
315
#' X <- matrix(rnorm(6),nrow=3,ncol=2,dimnames=list(c("A","B","C")))
#' Z <- matrix(rnorm(9),nrow=3,ncol=3,dimnames=list(c("B","A","B")))
#' match.samples(X,Z)
Rauschenberger's avatar
Rauschenberger committed
316
#' 
Rauschenberger's avatar
Rauschenberger committed
317
match.samples <- function(...,message=TRUE){
Rauschenberger's avatar
Rauschenberger committed
318
    
Rauschenberger's avatar
Rauschenberger committed
319
320
    list <- list(...)
    if(length(list)==1 & is.list(list[[1]])){list <- list[[1]]}
Rauschenberger's avatar
Rauschenberger committed
321
322
323
324
325
    if(is.null(names(list))){
        names(list) <- sapply(substitute(list(...))[-1],deparse)
    }
    names <- names(list)
    
Rauschenberger's avatar
Rauschenberger committed
326
    # check input
Rauschenberger's avatar
Rauschenberger committed
327
328
329
    cond <- sapply(list,function(x) !is.matrix(x))
    if(any(cond)){
        stop("Provide matrices!",call.=FALSE)
Rauschenberger's avatar
Rauschenberger committed
330
    }
Rauschenberger's avatar
Rauschenberger committed
331
332
333
    cond <- sapply(list,function(x) is.null(rownames(x)))
    if(any(cond)){
        stop("Provide row names!",call.=FALSE)
Rauschenberger's avatar
Rauschenberger committed
334
335
    }
    
Rauschenberger's avatar
Rauschenberger committed
336
    # remove duplicated samples
Rauschenberger's avatar
Rauschenberger committed
337
    duplic <- lapply(list,function(x) duplicated(rownames(x)))
Rauschenberger's avatar
Rauschenberger committed
338
    for(i in seq_along(list)){
Rauschenberger's avatar
Rauschenberger committed
339
340
341
        number <- round(100*mean(duplic[[i]]))
        if(message){message(number," duplicates in \"",names[i],"\"")}
        list[[i]] <- list[[i]][!duplic[[i]],,drop=FALSE]
Rauschenberger's avatar
Rauschenberger committed
342
    }
Rauschenberger's avatar
Rauschenberger committed
343
344
    
    # retain overlapping samples
Rauschenberger's avatar
Rauschenberger committed
345
346
347
    all <- Reduce(f=intersect,x=lapply(list,rownames))
    for(i in seq_along(list)){
        percent <- round(100*mean(rownames(list[[i]]) %in% all))
Rauschenberger's avatar
Rauschenberger committed
348
        if(message){message(percent,"% overlap in \"",names[i],"\"")}
Rauschenberger's avatar
Rauschenberger committed
349
        list[[i]] <- list[[i]][all,,drop=FALSE]
Rauschenberger's avatar
Rauschenberger committed
350
    }
Rauschenberger's avatar
Rauschenberger committed
351
352
    
    # check output
Rauschenberger's avatar
Rauschenberger committed
353
354
    cond <- sapply(list,function(x) any(duplicated(rownames(x))))
    if(any(cond)){
Rauschenberger's avatar
Rauschenberger committed
355
356
        stop("Duplicate samples!",call.=FALSE)
    }
Rauschenberger's avatar
Rauschenberger committed
357
358
    cond <- sapply(list,function(x) nrow(x)!=nrow(list[[1]]))
    if(any(cond)){
Rauschenberger's avatar
Rauschenberger committed
359
360
        stop("Different sample sizes!",call.=FALSE)
    }
Rauschenberger's avatar
Rauschenberger committed
361
362
    cond <- sapply(list,function(x) any(rownames(x)!=rownames(list[[1]])))
    if(any(cond)){
Rauschenberger's avatar
Rauschenberger committed
363
364
365
        stop("Different sample names!",call.=FALSE)
    }
    
Rauschenberger's avatar
Rauschenberger committed
366
    return(list)
Rauschenberger's avatar
Rauschenberger committed
367
368
369
370
371
372
373
}

#' @export
#' @title
#' Adjust library sizes
#' 
#' @description
Rauschenberger's avatar
Rauschenberger committed
374
#' This function adjusts RNA-seq expression data for different library sizes.
Rauschenberger's avatar
Rauschenberger committed
375
#' 
Rauschenberger's avatar
Rauschenberger committed
376
377
#' @param x
#' matrix with \eqn{n} rows (samples) and \eqn{p} columns (variables)
Rauschenberger's avatar
Rauschenberger committed
378
379
#' 
#' @examples
Rauschenberger's avatar
Rauschenberger committed
380
381
382
383
#' n <- 5; p <- 10
#' x <- matrix(rnbinom(n=n*p,mu=5,size=1/0.5),nrow=n,ncol=p)
#' x[1,] <- 10*x[1,]
#' adjust.samples(x)
Rauschenberger's avatar
Rauschenberger committed
384
#' 
Rauschenberger's avatar
Rauschenberger committed
385
adjust.samples <- function(x){
Rauschenberger's avatar
Rauschenberger committed
386
387
388
389
390
391
    if(!is.matrix(x)){
        stop("no matrix argument",call.=FALSE)
    }
    if(!is.numeric(x)){
        stop("no numeric argument",call.=FALSE)
    }
Rauschenberger's avatar
Rauschenberger committed
392
    if(!is.integer(x)&&any(round(x)!=x)){
Rauschenberger's avatar
Rauschenberger committed
393
        warning("non-integer values",call.=FALSE)
Rauschenberger's avatar
Rauschenberger committed
394
    }
Rauschenberger's avatar
Rauschenberger committed
395
    if(any(x<0)){
Rauschenberger's avatar
Rauschenberger committed
396
        warning("negative values",call.=FALSE)
Rauschenberger's avatar
Rauschenberger committed
397
    }
Rauschenberger's avatar
Rauschenberger committed
398
399
400
    n <- nrow(x); p <- ncol(x)
    lib.size <- rowSums(x)
    norm.factors <- edgeR::calcNormFactors(object=t(x),lib.size=lib.size)
Rauschenberger's avatar
Rauschenberger committed
401
    gamma <- norm.factors*lib.size/mean(lib.size)
Rauschenberger's avatar
Rauschenberger committed
402
    gamma <- matrix(gamma,nrow=n,ncol=p,byrow=FALSE)
Rauschenberger's avatar
Rauschenberger committed
403
404
    x <- x/gamma
    return(x)
Rauschenberger's avatar
Rauschenberger committed
405
406
407
408
409
410
411
412
413
}

#' @export
#' @title
#' Adjust exon length
#' 
#' @description
#' This function adjusts exon expression data for different exon lengths.
#' 
Rauschenberger's avatar
Rauschenberger committed
414
#' @param x
Rauschenberger's avatar
Rauschenberger committed
415
416
#' matrix with \eqn{n} rows (samples) and \eqn{p} columns (exons)
#' 
Rauschenberger's avatar
Rauschenberger committed
417
#' @param offset
Rauschenberger's avatar
Rauschenberger committed
418
#' exon length\strong{:} vector of length \eqn{p}
Rauschenberger's avatar
Rauschenberger committed
419
#' 
Rauschenberger's avatar
Rauschenberger committed
420
421
#' @param group
#' gene names\strong{:} vector of length \eqn{p}
Rauschenberger's avatar
Rauschenberger committed
422
423
424
425
#' 
#' @examples
#' NA
#' 
Rauschenberger's avatar
Rauschenberger committed
426
adjust.covariates <- function(x,offset,group){
Rauschenberger's avatar
Rauschenberger committed
427
428
429
    if(!is.numeric(x)|!is.matrix(x)){
        stop("Argument \"x\" is no numeric matrix.",call.=FALSE)
    }
Rauschenberger's avatar
Rauschenberger committed
430
431
432
433
434
    if(!is.numeric(offset)|!is.vector(offset)){
        stop("Argument \"offset\" is no numeric vector.",call.=FALSE)
    }
    if(any(offset<0)){
        stop("Argument \"offset\" takes negative values",call.=FALSE)   
Rauschenberger's avatar
Rauschenberger committed
435
    }
Rauschenberger's avatar
Rauschenberger committed
436
437
438
439
    if(!is.character(group)|!is.vector(group)){
        stop("Argument \"group\" is no character vector.",call.=FALSE)
    }
    if(ncol(x)!=length(group)|ncol(x)!=length(offset)){
Rauschenberger's avatar
Rauschenberger committed
440
441
        stop("Contradictory dimensions.",call.=FALSE)
    }
Rauschenberger's avatar
Rauschenberger committed
442
443
    n <- nrow(x); p <- ncol(x); names <- dimnames(x)
    x <- as.numeric(x)
Rauschenberger's avatar
Rauschenberger committed
444
    offset <- rep(offset,each=n)
Rauschenberger's avatar
Rauschenberger committed
445
446
447
448
449
450
451
    group <- strsplit(group,split=",")
    group <- sapply(group,function(x) x[[1]][1])
    group <- rep(group,each=n)
    lmer <- lme4::lmer(x ~ offset + (1|group))
    x <- matrix(stats::residuals(lmer),nrow=n,ncol=p,dimnames=names)
    x <- x-min(x)
    return(x)
Rauschenberger's avatar
Rauschenberger committed
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
}

#' @export
#' @title
#' Search for genes
#' 
#' @description
#' This function retrieves all genes on a chromosome.
#' 
#' @param chr
#' chromosome\strong{:} integer 1-22
#' 
#' @param path
#' path to gene transfer format files (.gtf)
#' 
#' @param release
#' character "NCBI36", "GRCh37", or "GRCh38"
#' 
#' @param build
#' integer 49-91
#' 
#' @examples
#' NA
#' 
Rauschenberger's avatar
Rauschenberger committed
476
477
478
map.genes <- function(chr,path=getwd(),release="GRCh37",build=71){
    
    # check input
Rauschenberger's avatar
Rauschenberger committed
479
    if(!chr %in% 1:22){
Rauschenberger's avatar
Rauschenberger committed
480
481
482
483
484
485
486
487
488
        stop("Invalid argument \"chr\".",call.=FALSE)
    }
    if(!release %in% c("NCBI36","GRCh37","GRCh38")){
        stop("Invalid argument \"release\".",call.=FALSE)
    }
    if(!build %in% 49:91){
        stop("Invalid argument \"build\".",call.=FALSE)
    }
    
Rauschenberger's avatar
Rauschenberger committed
489
490
    file <- paste0("Homo_sapiens.",release,".",build,".gtf")
    if(!file.exists(file.path(path,file))){
Rauschenberger's avatar
Rauschenberger committed
491
492
493
494
495
496
497
498
499
        url <- paste0("ftp://ftp.ensembl.org/pub/release-",build,
                      "/gtf/homo_sapiens/",file,".gz")
        destfile <- file.path(path,paste0(file,".gz"))
        utils::download.file(url=url,destfile=destfile,method="auto")
        R.utils::gunzip(filename=destfile,remove=FALSE,overwrite=TRUE)
    }
    object <- refGenome::ensemblGenome()
    refGenome::basedir(object) <- path
    refGenome::read.gtf(object,filename=file)
Rauschenberger's avatar
Rauschenberger committed
500
501
502
503
504
505
    x <- refGenome::getGenePositions(object=object,by="gene_id")
    x <- x[x$seqid==chr & x$gene_biotype=="protein_coding",]
    x <- x[,c("gene_id","seqid","start","end")]
    rownames(x) <- NULL
    colnames(x)[colnames(x)=="seqid"] <- "chr"
    return(x)
Rauschenberger's avatar
Rauschenberger committed
506
507
508
509
510
511
512
513
514
}

#' @export
#' @title
#' Search for exons
#' 
#' @description
#' This function
#' 
Rauschenberger's avatar
Rauschenberger committed
515
#' @param gene
Rauschenberger's avatar
Rauschenberger committed
516
517
#' gene names\strong{:} vector with one entry per gene,
#' including the gene names
Rauschenberger's avatar
Rauschenberger committed
518
#' 
Rauschenberger's avatar
Rauschenberger committed
519
#' @param exon
Rauschenberger's avatar
Rauschenberger committed
520
521
522
#' exon names\strong{:} vector with one entry per exon,
#' including the corresponding \emph{gene} names
#' (separated by comma if multiple gene names)
Rauschenberger's avatar
Rauschenberger committed
523
524
525
526
527
528
#' 
#' @details
#' The exon names should contain the gene names. For each gene, this function
#' returns the indices of the exons.
#' 
#' @examples
Rauschenberger's avatar
Rauschenberger committed
529
530
531
#' gene <- c("A","B","C")
#' exon <- c("A","A,B","B","B,C","C")
#' map.exons(gene,exon)
Rauschenberger's avatar
Rauschenberger committed
532
#'
Rauschenberger's avatar
Rauschenberger committed
533
534
535
map.exons <- function(gene,exon){
    p <- length(gene)
    x <- list()
Rauschenberger's avatar
Rauschenberger committed
536
537
538
    pb <- utils::txtProgressBar(min=0,max=p,style=3)
    for(i in seq_len(p)){
        utils::setTxtProgressBar(pb=pb,value=i)
Rauschenberger's avatar
Rauschenberger committed
539
540
        which <- as.integer(grep(pattern=gene[i],x=exon))
        x[[i]] <- which
Rauschenberger's avatar
Rauschenberger committed
541
    }
Rauschenberger's avatar
Rauschenberger committed
542
543
544
    close(con=pb)
    names(x) <- gene
    return(x)
Rauschenberger's avatar
Rauschenberger committed
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
}

#' @export
#' @title
#' Search for SNPs
#' 
#' @description
#' This function
#' 
#' @param gene.chr
#' chromosome\strong{:}
#' numeric vector with one entry per gene
#' 
#' @param gene.start
#' start position\strong{:}
#' numeric vector with one entry per gene
#' 
#' @param gene.end
#' end position\strong{:}
#' numeric vector with one entry per gene
#' 
#' @param snp.chr
#' integer 1-22
#' 
#' @param snp.pos
#' chromosomal position of SNPs\strong{:}
#' numeric vector with one entry per SNP
#' 
Rauschenberger's avatar
Rauschenberger committed
573
574
575
576
#' @param dist
#' number of base pairs before start position\strong{:}
#' integer
#' 
Rauschenberger's avatar
Rauschenberger committed
577
#' @examples
Rauschenberger's avatar
Rauschenberger committed
578
579
580
#' gene.chr <- rep(1,times=5)
#' gene.start <- 1:5
#' gene.end <- 2:6
Rauschenberger's avatar
Rauschenberger committed
581
#'
Rauschenberger's avatar
Rauschenberger committed
582
583
584
585
586
587
#' snp.chr <- rep(1,times=100)
#' snp.pos <- seq(from=1,to=4.9,length.out=100)
#' 
#' map.snps(gene.chr,gene.start,gene.end,snp.chr,snp.pos,dist=0)
#'
map.snps <- function(gene.chr,gene.start,gene.end,snp.chr,snp.pos,dist=10^3){
Rauschenberger's avatar
Rauschenberger committed
588
589
590
591
    if(length(gene.chr)!=length(gene.start)|length(gene.chr)!=length(gene.end)){
        stop("Invalid.",call.=FALSE)
    }
    p <- length(gene.start)
Rauschenberger's avatar
Rauschenberger committed
592
    x <- data.frame(from=integer(length=p),to=integer(length=p))
Rauschenberger's avatar
Rauschenberger committed
593
594
595
596
597
    pb <- utils::txtProgressBar(min=0,max=p,style=3)
    for(i in seq_len(p)){ # 
        utils::setTxtProgressBar(pb=pb,value=i)
        chr <- snp.chr == gene.chr[i]
        if(!any(chr)){next}
Rauschenberger's avatar
Rauschenberger committed
598
        start <- snp.pos >= (gene.start[i] - dist)
Rauschenberger's avatar
Rauschenberger committed
599
600
601
        end <- snp.pos <= gene.end[i] + 0
        which <- as.integer(which(chr & start & end))
        if(length(which)==0){next}
Rauschenberger's avatar
Rauschenberger committed
602
603
        x$from[i] <- min(which)
        x$to[i] <- max(which)
Rauschenberger's avatar
Rauschenberger committed
604
605
606
        if(length(which)==1){next}
        if(!all(diff(which)==1)){stop("SNPs are in wrong order!")}
    }
Rauschenberger's avatar
Rauschenberger committed
607
608
    close(con=pb)
    return(x)
Rauschenberger's avatar
Rauschenberger committed
609
610
611
612
}

#' @export
#' @title
Rauschenberger's avatar
Rauschenberger committed
613
#' Drop trivial tests
Rauschenberger's avatar
Rauschenberger committed
614
615
616
617
618
619
620
621
622
#' 
#' @description
#' This function
#' 
#' @param map
#' list with names "genes", "exons", and "snps"
#' (output from \code{map.genes}, \code{map.exons}, and \code{map.snps})
#' 
#' @details
Rauschenberger's avatar
Rauschenberger committed
623
#' This functions drops tests for genes without SNPs or with a single exon.
Rauschenberger's avatar
Rauschenberger committed
624
625
626
627
#' 
#' @examples
#' NA
#' 
Rauschenberger's avatar
Rauschenberger committed
628
drop.trivial <- function(map){
Rauschenberger's avatar
Rauschenberger committed
629
630
631
632
633
634
635
636
637
638
639
    
    # check input
    if(length(map)!=3){
        stop("Unexpected argument length.",call.=FALSE)
    }
    if(any(names(map)!=c("genes","exons","snps"))){
        stop("Unexpected argument names.",call.=FALSE)
    }
    
    # search
    p <- nrow(map$genes)
Rauschenberger's avatar
Rauschenberger committed
640
641
    pass <- rep(NA,times=p)
    pb <- utils::txtProgressBar(min=0,max=p,style=3)
Rauschenberger's avatar
Rauschenberger committed
642
    for(i in seq_len(p)){
Rauschenberger's avatar
Rauschenberger committed
643
644
645
646
647
648
649
650
651
652
        utils::setTxtProgressBar(pb=pb,value=i)
        ys <- map$exons[[i]]
        check <- logical()
        # Exclude genes without SNPs:
        check[1] <- map$snps$from[i] > 0
        check[2] <- map$snps$to[i] > 0
        # Exclude genes with single exon:
        check[3] <- length(ys) > 1
        pass[i] <- all(check)
    }
Rauschenberger's avatar
Rauschenberger committed
653
    close(con=pb)
Rauschenberger's avatar
Rauschenberger committed
654
655
656
657
658
659
660
661
662
    
    # check output
    if(any(pass[map$snps$to==0 & map$snps$from==0])){
        stop("Genes without any SNPs.",call.=FALSE)
    }
    if(any(pass[sapply(map$exons,length)<2])){
        stop("Genes without multiple exons.",call.=FALSE)
    }
    
Rauschenberger's avatar
Rauschenberger committed
663
664
665
666
    map$genes <- map$genes[pass,]
    map$exons <- map$exons[pass]
    map$snps <- map$snps[pass,]
    return(map)
Rauschenberger's avatar
Rauschenberger committed
667
668
669
670
671
}


#' @export
#' @title
Rauschenberger's avatar
Rauschenberger committed
672
#' Conduct single tests
Rauschenberger's avatar
Rauschenberger committed
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
#' 
#' @description
#' This function
#' 
#' @param Y
#' exon expression\strong{:}
#' matrix with \eqn{n} rows (samples) and \eqn{p} columns (exons)
#' 
#' @param X
#' SNP genotype\strong{:}
#' matrix with \eqn{n} rows (samples) and \eqn{q} columns (SNPs)
#' 
#' @param map
#' list with names "genes", "exons", and "snps"
#' (output from \code{map.genes}, \code{map.exons}, and \code{map.snps})
#' 
#' @param i
#' gene index\strong{:}
#' integer between \eqn{1} and \code{nrow(map$genes)}
#' 
#' @param limit
#' cutoff for rounding \code{p}-values
#' 
#' @param steps
#' size of permutation chunks\strong{:}
#' integer vector
#' 
Rauschenberger's avatar
Rauschenberger committed
700
701
702
#' @param rho
#' correlation\strong{:}
#' numeric vector with values between \eqn{0} and \eqn{1}
Rauschenberger's avatar
Rauschenberger committed
703
#' 
Rauschenberger's avatar
Rauschenberger committed
704
705
706
707
708
709
710
711
#' @details
#' The maximum number of permutations equals \code{sum(steps)}. Permutations is
#' interrupted if at least \code{limit} test statistics for the permuted data
#' are larger than the test statistic for the observed data.
#' 
#' @examples
#' NA
#' 
Rauschenberger's avatar
Rauschenberger committed
712
test.single <- function(Y,X,map,i,limit=NULL,steps=NULL,rho=c(0,0.5,1)){
Rauschenberger's avatar
Rauschenberger committed
713
714
715
    
    if(is.null(limit)){limit <- 5}
    if(is.null(steps)){steps <- c(10,20,20,50)}
Rauschenberger's avatar
Rauschenberger committed
716
    
Rauschenberger's avatar
Rauschenberger committed
717
    # check input
Rauschenberger's avatar
Rauschenberger committed
718
719
720
721
722
723
    if(!is.numeric(limit)){
        stop("Argument \"limit\" is not numeric.",call.=FALSE)
    }
    if(limit<1){
        stop("Argument \"limit\" is below one.",call.=FALSE)
    }
Rauschenberger's avatar
Rauschenberger committed
724
    if(!is.numeric(steps)|!is.vector(steps)){
Rauschenberger's avatar
Rauschenberger committed
725
726
727
728
729
730
        stop("Argument \"steps\" is no numeric vector.",call.=FALSE)
    }
    if(sum(steps)<2){
        stop("Too few permutations \"sum(steps)\".",call.=FALSE)
    }
    
Rauschenberger's avatar
Rauschenberger committed
731
    # extract data
Rauschenberger's avatar
Rauschenberger committed
732
    ys <- map$exons[[i]]
Rauschenberger's avatar
Rauschenberger committed
733
    y <- Y[,ys,drop=FALSE]
Rauschenberger's avatar
Rauschenberger committed
734
    xs <- seq(from=map$snps$from[i],to=map$snps$to[i],by=1)
Rauschenberger's avatar
Rauschenberger committed
735
    x <- X[,xs,drop=FALSE]
Rauschenberger's avatar
Rauschenberger committed
736
    
Rauschenberger's avatar
Rauschenberger committed
737
    # test effects
Rauschenberger's avatar
Rauschenberger committed
738
739
    pvalue <- rep(x=NA,times=length(rho))
    for(j in seq_along(rho)){
Rauschenberger's avatar
Rauschenberger committed
740
741
        tstat <- spliceQTL:::G2.multin(
            dep.data=y,indep.data=x,nperm=steps[1]-1,rho=rho[j])$Sg
Rauschenberger's avatar
Rauschenberger committed
742
        for(nperm in steps[-1]){
Rauschenberger's avatar
Rauschenberger committed
743
744
            tstat <- c(tstat,spliceQTL:::G2.multin(
                dep.data=y,indep.data=x,nperm=nperm,rho=rho[j])$Sg[-1])
Rauschenberger's avatar
Rauschenberger committed
745
            if(sum(tstat >= tstat[1]) >= limit){break}
Rauschenberger's avatar
Rauschenberger committed
746
        }
Rauschenberger's avatar
Rauschenberger committed
747
        pvalue[j] <- mean(tstat >= tstat[1],na.rm=TRUE)
Rauschenberger's avatar
Rauschenberger committed
748
    }
Rauschenberger's avatar
Rauschenberger committed
749

Rauschenberger's avatar
Rauschenberger committed
750
751
752
753
    return(pvalue)
}


Rauschenberger's avatar
Rauschenberger committed
754
755
#' @export
#' @title
Rauschenberger's avatar
Rauschenberger committed
756
#' Conduct multiple tests
Rauschenberger's avatar
Rauschenberger committed
757
758
#' 
#' @description
Rauschenberger's avatar
Rauschenberger committed
759
#' This function ...
Rauschenberger's avatar
Rauschenberger committed
760
761
762
763
764
765
766
767
768
#' 
#' @param Y
#' exon expression\strong{:}
#' matrix with \eqn{n} rows (samples) and \eqn{p} columns (exons)
#' 
#' @param X
#' SNP genotype\strong{:}
#' matrix with \eqn{n} rows (samples) and \eqn{q} columns (SNPs)
#' 
Rauschenberger's avatar
Rauschenberger committed
769
770
771
772
#' @param map
#' list with names "genes", "exons", and "snps"
#' (output from \code{map.genes}, \code{map.exons}, and \code{map.snps})
#' 
Rauschenberger's avatar
Rauschenberger committed
773
774
775
776
777
778
779
780
781
782
783
#' @param rho
#' correlation\strong{:}
#' numeric vector with values between \eqn{0} and \eqn{1}
#' 
#' @details
#' Automatic adjustment of the number of permutations
#' such that Bonferroni-significant p-values are possible.
#' 
#' @examples
#' NA
#' 
Rauschenberger's avatar
Rauschenberger committed
784
785
test.multiple <- function(Y,X,map,rho=c(0,0.5,1)){
    
Rauschenberger's avatar
Rauschenberger committed
786
787
788
    p <- nrow(map$genes)
    
    # permutations
Rauschenberger's avatar
Rauschenberger committed
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
    if(FALSE){
        min <- 5
        max <- p/0.05+1
        limit <- ceiling(0.05*max/p)
        base <- 1.5 # adjust sequence
        from <- log(min,base=base)
        to <- log(max,base=base)
        steps <- c(min,diff(unique(round(base^(seq(from=from,to=to,length.out=20))))))
    }
    
    if(TRUE){
        max <- p/0.05+1
        limit <- ceiling(0.05*max/p)
        steps <- diff(limit^seq(from=1,to=log(max)/log(limit),length.out=pmin(p,20)))
        steps <- c(limit,round(steps))
        steps[length(steps)] <- max-sum(steps[-length(steps)])
    }
Rauschenberger's avatar
Rauschenberger committed
806
    
Rauschenberger's avatar
Rauschenberger committed
807
    if(max != sum(steps)){stop("Invalid combination?",call.=FALSE)}
Rauschenberger's avatar
Rauschenberger committed
808
809
810
811
812
    
    # parallel computation
    type <- ifelse(test=.Platform$OS.type=="windows",yes="PSOCK",no="FORK")
    cluster <- parallel::makeCluster(spec=8,type=type)
    parallel::clusterSetRNGStream(cl=cluster,iseed=1)
Rauschenberger's avatar
Rauschenberger committed
813
    parallel::clusterExport(cl=cluster,varlist=c("Y","X","map","limit","steps","rho"),envir=environment())
Rauschenberger's avatar
Rauschenberger committed
814
    start <- Sys.time()
Rauschenberger's avatar
Rauschenberger committed
815
    pvalue <- parallel::parLapply(cl=cluster,X=seq_len(p),fun=function(i) spliceQTL::test.single(Y=Y,X=X,map=map,i=i,limit=limit,steps=steps,rho=rho))
Rauschenberger's avatar
Rauschenberger committed
816
817
818
819
820
    end <- Sys.time()
    parallel::stopCluster(cluster)
    rm(cluster)
    
    # tyding up
Rauschenberger's avatar
Rauschenberger committed
821
    pvalue <- do.call(what=rbind,args=pvalue)
Rauschenberger's avatar
Rauschenberger committed
822
    colnames(pvalue) <- paste0("rho=",rho)
Rauschenberger's avatar
Rauschenberger committed
823
824
    rownames(pvalue) <- map$genes$gene_id
    
Rauschenberger's avatar
Rauschenberger committed
825
    return(pvalue)
Rauschenberger's avatar
Rauschenberger committed
826
827
828
829
}



Rauschenberger's avatar
Rauschenberger committed
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
#--- spliceQTL test functions --------------------------------------------------

# Function: G2.multin
# This is to compute the G2 test statistic under the assumption that the response follows a multinomial distribution
### Input 
### dep data and indep data with samples on the rows and genes on the columns
### grouping: Either a logical value = F or a matrix with a single column and same number of rows as samples. 
###         Column name should be defined.
###         Contains clinical information of the samples. 
###         Should have two groups only. 
### nperm : number of permutations 
### rho: the null correlation between SNPs
### mu: the null correlation between observations corresponding to different exons and different individuals

### Output
### A list containing G2 p.values and G2 test statistics

### Example : G2T = G2(dep.data = cgh, indep.data = expr, grouping=F, stand=TRUE, nperm=1000)
### G2 p.values : G2T$G2p
### G2 TS : G2T$$Sg

Rauschenberger's avatar
Rauschenberger committed
851
G2.multin <- function(dep.data,indep.data,stand=TRUE,nperm=100,grouping=F,rho=0,mu=0){
Rauschenberger's avatar
Rauschenberger committed
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
    
    nperm = nperm
    ## check for the number of samples in dep and indep data
    
    
    if (nrow(dep.data)!=nrow(indep.data)){
        cat("number of samples not same in dep and indep data","\n")
    }
    
    if(any(abs(rho)>1)){
        cat("correlations rho larger than abs(1) are not allowed")
    }
    
    nresponses <- ncol(dep.data)
    ncovariates <- ncol(indep.data)
    ### centering and standardizing the data are not done in this case
    
    #  dep.data = scale(dep.data,center=T,scale=stand)
    #  indep.data = scale(indep.data,center=T,scale=stand)
    
    #### No  grouping of the samples.
    
    ## Calculate U=(I-H)Y and UU', where Y has observations on rows; also tau.mat=X*W.rho*X', 
    ##   where X has observations on rows and variables on columns
    ##  and W.rho = I + rho*(J-I), a square matrix with as many rows as columns in X
    ## NOTE: this formulation uses X with n obs on the rows and m covariates no the columns, so it is the transpose of the first calculations
    nsamples <- nrow(dep.data)
    n.persample <- rowSums(dep.data)
    n.all <- sum(dep.data)
    H <- (1/n.all)*matrix( rep(n.persample,each=nsamples),nrow=nsamples,byrow=T)
    U <- (diag(rep(1,nsamples)) - H) %*% dep.data
    ## Now we may have a vector of values for rho - so we define tau.mat as an array, with the 3rd index corresponding to the value of rho
    tau.mat <- array(0,dim=c(nsamples,nsamples,length(rho)))
    for(xk in 1:length(rho))  
    {  
        if (rho[xk]==0) { tau.mat[,,xk] <- tcrossprod(indep.data) } 
        else { w.rho <- diag(rep(1,ncovariates)) + rho[xk]*(tcrossprod(rep(1,ncovariates)) -diag(rep(1,ncovariates))  )
        tau.mat[,,xk] <- indep.data %*% w.rho %*% t(indep.data)}
        
    }
    ######################################
    ### NOTES ARMIN START ################
    # all(X %*% t(X) == tau.mat[,,1]) # rho = 0 -> TRUE
    # all(X %*% (t(X) %*% X) %*% t(X) == tau.mat[,,1]) # rho = 1
    # plot(as.numeric(X %*% (t(X) %*% X) %*% t(X)),as.numeric(tau.mat[,,1]))
    ### NOTES ARMIN END ##################
    ######################################
    samp_names = 1:nsamples ## this was rownames(indep.data), but I now do this so that rownames do not have to be added to the array tau.mat
    Sg = get.g2stat.multin(U,mu=mu,rho=rho,tau.mat)
    ### now we will have a vector as result, with one value per combination of values of rho and mu
    #
    ### G2 
    ### Permutations
    # When using permutations: only the rows of tau.mat are permuted
    # To check how the permutations can be efficiently applied, see tests_permutation_g2_multin.R
    
    
    perm_samp = matrix(0, nrow=nrow(indep.data), ncol=nperm)   ## generate the permutation matrix
    for(i in 1:ncol(perm_samp)){
        perm_samp[,i] = samp_names[sample(1:length(samp_names),length(samp_names))]
    }
    
    ## permutation starts - recompute tau.mat  (or recompute U each time)
    for (perm in 1:nperm){
        tau.mat.perm = tau.mat[perm_samp[,perm],,,drop=FALSE]          # permute rows
        tau.mat.perm = tau.mat.perm[,perm_samp[,perm],,drop=FALSE]     # permute columns
        
Rauschenberger's avatar
Rauschenberger committed
919
        Sg = c(Sg,spliceQTL:::get.g2stat.multin(U, mu=mu,rho=rho,tau.mat.perm) )
Rauschenberger's avatar
Rauschenberger committed
920
921
922
923
924
925
926
927
928
929
930
931
    }
    
    
    ########################################################################
    
    #### G2 test statistic
    # *** recompute for a vector of values for each case - just reformat the result with as many rows as permutations + 1,
    # and as many columns as combinations of values of rho and mu
    Sg = matrix(Sg,nrow=nperm+1,ncol=length(mu)*length(rho))
    colnames(Sg) <- paste(rep("rho",ncol(Sg)),rep(1:length(rho),each=length(mu)),rep("mu",ncol(Sg)),rep(1:length(mu),length(rho)) )
    
    ### Calculte G2 pval
Rauschenberger's avatar
Rauschenberger committed
932
    G2p =  apply(Sg,2,spliceQTL:::get.pval.percol) 
Rauschenberger's avatar
Rauschenberger committed
933
934
935
936
937
938
939
940
941
942
943
944
    
    return (list(perm = perm_samp,G2p = G2p,Sg = Sg))
}

# Function: get.g2stat.multin
# Computes the G2 test statistic given two data matrices, under a multinomial distribution
# This is used internally by the G2 function
# Inputs: 
#  U = (I-H)Y, a n*K matrix where n=number obs and K=number multinomial responses possible
#  tau.mat = X' W.rho X, a n*n matrix : both square, symmetric matrices with an equal number of rows
# Output: test statistic (single value)
# 
Rauschenberger's avatar
Rauschenberger committed
945
get.g2stat.multin <- function(U, mu, rho, tau.mat){
Rauschenberger's avatar
Rauschenberger committed
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
    g2tstat <- NULL
    for(xk in 1:length(rho))
    {
        for(xj in 1:length(mu))
        {
            if(mu[xj]==0) { g2tstat <- c(g2tstat, sum( diag( tcrossprod(U) %*% tau.mat[,,xk] ) ) )
            } else {
                g2tstat <- c(g2tstat, (1-mu[xj])*sum(diag( tcrossprod(U) %*% tau.mat[,,xk] ) ) + mu[xj]*sum( t(U) %*% tau.mat[,,xk] %*% U )  )
            }
            
        }
    }
    g2tstat
}

# Function: get.pval.percol
# This function takes a vector containing the observed test stat as the first entry, followed by values generated by permutation,
# and computed the estimated p-value
# Input
# x: a vector with length nperm+1
# Output
# the pvalue computed
get.pval.percol <- function(x){
    pval = mean(x[1]<= c(Inf , x[2:length(x)]))
    pval
}