-
Notifications
You must be signed in to change notification settings - Fork 13
/
assocTestSingle.R
203 lines (157 loc) · 8.18 KB
/
assocTestSingle.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
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
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
setGeneric("assocTestSingle", function(gdsobj, ...) standardGeneric("assocTestSingle"))
## do we want the GxE.return.cov option?
## do we want to make imputing to the mean optional?
setMethod("assocTestSingle",
"SeqVarIterator",
function(gdsobj, null.model, test=c("Score", "Score.SPA", "BinomiRare", "CMP"),
recalc.pval.thresh=0.05, fast.score.SE=FALSE, GxE=NULL,
geno.coding=c("additive", "dominant", "recessive"),
sparse=TRUE, imputed=FALSE, male.diploid=TRUE, genome.build=c("hg19", "hg38"),
BPPARAM=bpparam(), verbose=TRUE) {
test <- match.arg(test)
geno.coding <- match.arg(geno.coding)
# don't use sparse matrices for imputed dosages
if (imputed) sparse <- FALSE
# Convert old null model format if necessary.
null.model <- .updateNullModelFormat(null.model)
# check that the provided null model is compatible with the requested test
.checkNullModelTestSingle(null.model = null.model, test = test,
recalc.pval.thresh = recalc.pval.thresh, fast.score.SE = fast.score.SE, GxE = GxE)
# coerce null.model if necessary
if (sparse) null.model <- .nullModelAsMatrix(null.model)
# filter samples to match null model
sample.index <- .setFilterNullModel(gdsobj, null.model, verbose=verbose)
# get sex for calculating allele freq
sex <- validateSex(gdsobj)[sample.index]
if (!is.null(GxE)) GxE <- .modelMatrixColumns(null.model, GxE)
# check ploidy
if (SeqVarTools:::.ploidy(gdsobj) == 1) male.diploid <- FALSE
# results
#n.iter <- length(variantFilter(gdsobj))
#set.messages <- ceiling(n.iter / 100) # max messages = 100
if(verbose) message('Using ', bpnworkers(BPPARAM), ' CPU cores')
i <- 1
ITER <- function() {
iterate <- TRUE
if (i > 1) {
iterate <- iterateFilter(gdsobj, verbose=FALSE)
}
i <<- i + 1
if (!iterate) {
return(NULL)
}
var.info <- variantInfo(gdsobj, alleles=FALSE, expanded=TRUE)
if (!imputed) {
geno <- expandedAltDosage(gdsobj, use.names=FALSE, sparse=sparse)[sample.index,,drop=FALSE]
} else {
geno <- imputedDosage(gdsobj, use.names=FALSE)[sample.index,,drop=FALSE]
}
chr <- chromWithPAR(gdsobj, genome.build=genome.build)
return(list(var.info=var.info, geno=geno, chr=chr))
}
res <- bpiterate(ITER, .testGenoBlockSingle, BPPARAM=BPPARAM,
sex=sex, null.model=null.model, test=test,
recalc.pval.thresh=recalc.pval.thresh,
fast.score.SE=fast.score.SE, GxE=GxE,
geno.coding=geno.coding,
sparse=sparse, imputed=imputed,
male.diploid=male.diploid)
.stopOnError(res)
as.data.frame(rbindlist(res))
})
setMethod("assocTestSingle",
"GenotypeIterator",
function(gdsobj, null.model, test=c("Score", "Score.SPA", "BinomiRare", "CMP"),
recalc.pval.thresh=0.05, GxE=NULL,
geno.coding=c("additive", "dominant", "recessive"),
male.diploid=TRUE, BPPARAM=bpparam(), verbose=TRUE) {
test <- match.arg(test)
geno.coding <- match.arg(geno.coding)
# Convert old null model format if necessary.
null.model <- .updateNullModelFormat(null.model)
# check that the provided null model is compatible with the requested test
.checkNullModelTestSingle(null.model = null.model, test = test,
recalc.pval.thresh = recalc.pval.thresh, fast.score.SE = FALSE, GxE = GxE)
# filter samples to match null model
sample.index <- .sampleIndexNullModel(gdsobj, null.model)
# get sex for calculating allele freq
sex <- validateSex(gdsobj)[sample.index]
if (!is.null(GxE)) GxE <- .modelMatrixColumns(null.model, GxE)
# results
# n.iter <- length(snpFilter(gdsobj))
# set.messages <- ceiling(n.iter / 100) # max messages = 100
if(verbose) message('Using ', bpnworkers(BPPARAM), ' CPU cores')
i <- 1
ITER <- function() {
iterate <- TRUE
if (i > 1) {
iterate <- GWASTools::iterateFilter(gdsobj)
}
i <<- i + 1
if (!iterate) {
return(NULL)
}
var.info <- variantInfo(gdsobj)
geno <- getGenotypeSelection(gdsobj, scan=sample.index, order="selection",
transpose=TRUE, use.names=FALSE, drop=FALSE)
return(list(var.info=var.info, geno=geno, chr=var.info$chr))
}
res <- bpiterate(ITER, .testGenoBlockSingle, BPPARAM=BPPARAM,
sex=sex, null.model=null.model, test=test,
recalc.pval.thresh=recalc.pval.thresh,
fast.score.SE=FALSE, GxE=GxE,
geno.coding=geno.coding,
sparse=FALSE, imputed=FALSE,
male.diploid=male.diploid)
.stopOnError(res)
as.data.frame(rbindlist(res))
})
# function to process a block of genotype data
.testGenoBlockSingle <- function(x, sex, null.model, test,
recalc.pval.thresh, fast.score.SE, GxE,
geno.coding,
sparse, imputed, male.diploid, ...) {
# for BinomiRare and CMP, restrict to variants where the alternate allele is minor
if (test %in% c("BinomiRare", "CMP")) {
AF.max <- 0.5
} else {
AF.max <- 1
}
x <- .prepGenoBlock(x, AF.max=AF.max, geno.coding=geno.coding, imputed=imputed,
sex=sex, male.diploid=male.diploid)
var.info <- x$var.info
n.obs <- x$n.obs
freq <- x$freq
geno <- x$geno
rm(x)
# mean impute missing values
if (any(n.obs < nrow(geno))) {
geno <- .meanImpute(geno, freq$freq)
}
# do the test
if (ncol(geno) == 0){
res.i <- NULL
} else {
assoc <- testGenoSingleVar(null.model, G=geno, E=GxE, test=test,
recalc.pval.thresh=recalc.pval.thresh,
fast.score.SE=fast.score.SE)
res.i <- cbind(var.info, n.obs, freq, assoc)
}
#if (verbose & n.iter > 1 & i %% set.messages == 0) {
# message(paste("Iteration", i , "of", n.iter, "completed"))
#}
return(res.i)
}
# check that the provided null model is compatible with the requested test
.checkNullModelTestSingle <- function(null.model, test, recalc.pval.thresh, fast.score.SE, GxE){
calc.score <- test %in% c("Score", "Score.SPA") | (recalc.pval.thresh < 1)
if(fast.score.SE && !isNullModelFastScore(null.model)){
stop("null.model must have se.correction when fast.score.SE = TRUE; re-fit your null.model using `fitNullModelFastScore` or update your null.model using `nullModelFastScore`")
}
if(calc.score && !(fast.score.SE) && isNullModelSmall(null.model)){
stop("small null.model cannot be used with test options provided")
}
if(!is.null(GxE) && isNullModelSmall(null.model)){
stop("small null.model cannot be used with GxE")
}
}