/
beagle-false-positives.R
254 lines (227 loc) · 15.6 KB
/
beagle-false-positives.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
204
205
206
207
208
209
210
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
# Written 2013 by Peter Ralph and Graham Coop
#
# contact: petrel.harp@gmail.com
#
# To the extent possible under law, the author(s) have dedicated all copyright and related and neighboring rights to this software to the public domain worldwide. This software is distributed without any warranty.
#
# You should have received a copy of the CC0 Public Domain Dedication along with this software. If not, see <http://creativecommons.org/publicdomain/zero/1.0/>.
#
# Usage:
# Find false positive rate from beagle runs on randomly mixed up chromosomes.
source("ibd-blocks-fns.R")
# Load in false positives
if (!file.exists("false-pos-everything.Rdata")) {
chromlist <- c(1,13,22)
fpblocks <- do.call( rbind, lapply( chromlist, function (chrom) {
fp <- readblocks(paste("RANDOM.REGIONAL.POPRES_chr",chrom,".combined.fibd.gz",sep=""), chr=chrom, header=TRUE, endblocks=FALSE)
fp$chrom <- chrom
return(fp)
} ) )
save(fpblocks, chromlist, file="false-pos-everything.Rdata")
} else {
load("false-pos-everything.Rdata")
}
# Observed blocks
blocks <- do.call( rbind, lapply( chromlist, function (chrom) getblocks(chrom,"combined") ) )
# Restrict to those seen in at least two runs:
fpgb <- (fpblocks$nsegs > 1) & (fpblocks$score <= 1e-9) & (!is.na(fpblocks$maplen))
gb <- (blocks$nsegs > 1) & (blocks$score <= 1e-9)
require(hexbin)
with(fpblocks, plot( hexbin( log10(score) ~ log10(maplen) ), style="centroids" ) )
# How many FP at different score cutoffs and for different numbers of runs
png(file="chr22-nfp.png",width=5*72,height=5*72,res=72)
scuts <- exp( seq(log(1e-10),log(1e-15),len=100) )
nfp <- sapply(scuts, function (cutoff) sum( fpblocks$score<=cutoff, na.rm=TRUE ) )
npairs <- choose( length( unique( c(fpblocks$id1, fpblocks$id2) ) ), 2 )
nfp <- nfp/npairs
plot(scuts, nfp, log="x", type="l", xlab="score cutoff", ylab="number of false positives per pair", ylim=c(0,max(nfp)))
# by number of runs cutoff
for (k in 1:5) {
nfp <- sapply(scuts, function (cutoff) sum( (fpblocks$nsegs>k & fpblocks$score<=cutoff)/npairs, na.rm=TRUE ) )
lines(scuts, nfp, col=rainbow(5)[k], lty=2)
}
# How many FP at different score cutoffs and for different lengths
lcuts <- (1:5)-0.5
# by number of runs cutoff
for (k in 1:5) {
nfp <- sapply(scuts, function (cutoff) sum( (fpblocks$maplen>lcuts[k] & fpblocks$score<=cutoff & fpblocks$nsegs>1)/npairs, na.rm=TRUE ) )
lines(scuts, nfp, col=rainbow(5)[k], lty=1)
}
legend("topleft", col=rainbow(5), lty=c(rep(2,5),0,rep(1,5)), legend=c(paste("seen in at least", 1:5, "runs"),"seen in at least 2 runs and...", paste("at least", lcuts, "cM long")), cex=0.5)
dev.off()
# Tabulate this.
fptab <- with(fpblocks, table( pmin(15,floor(-log10(score))), nsegs, floor(maplen) ) )
names(dimnames(fptab)) <- c("log10score","nsegs","maplenCM")
ftable(addmargins(fptab))
# Small table
smfptab <- with(fpblocks, table( score <= 1e-10, nsegs>1, pmin(4,floor(maplen)) ) )
dimnames(smfptab) <- list(score=c(">1e-10","<=1e-10"),nsegs=c("1",">1"),maplenMb=c(paste(c(0:3),"Mb"),">4Mb") )
ftable(addmargins(smfptab))
# Plot these:
png(file="chr22-pos-vs-score-with-fp.png", width=5*72, height=5*72, res=72)
with(blocks[gb,], plot( maplen, score, log="xy", pch=20, col=adjustcolor("black",0.2), ylim=c(1e-23,1e-10), xlim=c(.5,10) ) )
with(fpblocks[fpgb,], points( maplen, score, pch=20, col=adjustcolor("red",0.2) ) )
legend("bottomleft",col=c("black","red"),legend=c("observed","falsepos"),pch=20)
dev.off()
# Plot proportions
png(file="chr22-observed-rates.png", width=8*72, height=5*72, res=72)
with(blocks[gb,], plot( ecdf(log10(maplen)) ) )
with(fpblocks[fpgb,], lines( ecdf(log10(maplen)), col="red" ) )
lines( seq(-2,1,len=50), sapply( 10^seq(-2,1,len=50), function (x) { sum( fpgb & fpblocks$maplen>x, na.rm=TRUE ) / sum( gb & blocks$maplen>x, na.rm=TRUE ) } ), col="green" )
legend("topleft", col=c("black","red","green"), legend=c("observed ecdf","falsepos ecdf","(num fp >x)/(num obs >x)"), lty=1)
dev.off()
# Examine position along the chromosome
png(file="chr22-fp-locations.png", width=8*72, height=6*72, res=72)
with( blocks[gb,], plot( mapmid, maplen, pch=20, col=adjustcolor("black",.3)) )
with( fpblocks[fpgb,], points( mapmid, maplen, pch=20, col=adjustcolor("red",.3) ) )
rug( chrmap$map, col=adjustcolor("black"))
legend("topleft",col=c("black","red"), pch=20, legend=c("observed","false pos"))
dev.off()
# Plot some examples
png(file="300-examples.png", width=7*72, height=5*72, res=72)
plotmany( unique( sample(fpblocks$id1,300)), fpblocks[fpgb,], chrom=22, lwd=3 )
dev.off()
#########
# FP rate by country pair
# False positives
if (!file.exists("false-pos-winnowed.Rdata")) {
chromlist <- c(1,13,22)
fpblocks <- do.call( rbind, lapply( chromlist, function (chrom) {
fp <- readblocks(paste("RANDOM.REGIONAL.POPRES_chr",chrom,".combined.winnowed.fibd.gz",sep=""), chr=chrom, header=TRUE, endblocks=FALSE)
fp$chrom <- chrom
return(fp)
} ) )
indivinfo <- getsampleinfo(remove.qc=TRUE)
fpblocks <- subset(fpblocks, (!is.na(maplen)) & indivinfo$YESOK[match(id1,indivinfo$SUBJID)] & indivinfo$YESOK[match(id2,indivinfo$SUBJID)] ) # remove qc'ed indivs
fp.ids <- unique( c(fpblocks$id1, fpblocks$id2) )
indivinfo <- droplevels( subset( indivinfo, SUBJID %in% fp.ids ) )
# We rearranged genomes before splitting out England, Wales, etc from UK
# check this by untarring e.g. reorder.REGIONAL.POPRES_chr13.beagle.tar.gz to test and running:
# tmpindivinfo <- getsampleinfo(remove.qc=FALSE)
# fnames <- list.files("test","^country.*",full.names=TRUE); z <- do.call( rbind, lapply( fnames, function (x) read.table(x,header=FALSE,stringsAsFactors=FALSE) ) );
# names(z) <- c("SUBJID","COUNTRY_SELF"); z$other <- as.character( indivinfo$COUNTRY_SELF[ match(z$SUBJID,indivinfo$SUBJID) ] );
# z <- subset(z,tmpindivinfo$YESOK[match(SUBJID,tmpindivinfo$SUBJID)]);
# subset(z,COUNTRY_SELF!=other); all( sort(unique(z$SUBJID)) == sort(fp.ids) ) # TRUE
for (x in list(
list( "United Kingdom", c("England","Scotland","Wales") ),
list( "Switzerland", c("Swiss French","Swiss German") ),
list( "Yugoslavia", c("Albania","Bosnia","Croatia","Serbia","Macedonia","Kosovo") )
) ) {
levels(indivinfo$COUNTRY_SELF)[levels(indivinfo$COUNTRY_SELF) %in% x[[2]]] <- x[[1]]
}
indivinfo$COUNTRY_SELF <- ordered(indivinfo$COUNTRY_SELF,levels=sort(levels(indivinfo$COUNTRY_SELF)))
# subset( as.data.frame(with(subset(indivinfo,SUBJID %in% fp.ids), table(COUNTRY_SELF,ORIG_COUNTRYSELF))), Freq>0 )
fpblocks$country1 <- ordered( indivinfo$COUNTRY_SELF[ match(fpblocks$id1,indivinfo$SUBJID) ], levels=(levels(indivinfo$COUNTRY_SELF)) )
fpblocks$country2 <- ordered( indivinfo$COUNTRY_SELF[ match(fpblocks$id2,indivinfo$SUBJID) ], levels=(levels(indivinfo$COUNTRY_SELF)) )
fpblocks$countrypair <- gsub(" ",".",with(fpblocks, paste( levels(fpblocks$country1)[ifelse(country1<country2,country1,country2)], levels(fpblocks$country1)[ifelse(country1<country2,country2,country1)], sep="-" ) ) )
# Our observed data
obsblocks <- subset( do.call( rbind, lapply( chromlist, function (chrom) getblocks(chrom,"winnowed") ) ), (id1 %in% fp.ids) & (id2 %in% fp.ids) )
obsblocks$country1 <- ordered( indivinfo$COUNTRY_SELF[ match(obsblocks$id1,indivinfo$SUBJID) ], levels=(levels(indivinfo$COUNTRY_SELF)) )
obsblocks$country2 <- ordered( indivinfo$COUNTRY_SELF[ match(obsblocks$id2,indivinfo$SUBJID) ], levels=(levels(indivinfo$COUNTRY_SELF)) )
obsblocks$countrypair <- gsub(" ",".",with(obsblocks, paste( levels(obsblocks$country1)[ifelse(country1<country2,country1,country2)], levels(obsblocks$country1)[ifelse(country1<country2,country2,country1)], sep="-" ) ) )
save(chromlist, indivinfo, fpblocks, obsblocks, file="false-pos-winnowed.Rdata")
} else {
load("false-pos-winnowed.Rdata")
}
# number of pairs
npairs <- countpairs(sampleinfo=indivinfo)
totalpairs <- sum(npairs)
# FP rate by length
lenbins <- quantile( obsblocks$maplen[obsblocks$maplen>0.0], seq(0,1,length.out=10), na.rm=TRUE )
midbins <- lenbins[-1]-diff(lenbins)/2; midbins[length(midbins)] <- midbins[length(midbins)-1]+3
fp.counts <- with(fpblocks, table( cut( maplen[maplen>0.0], breaks=lenbins ) ) )
true.counts <- with(obsblocks, table( cut( maplen[maplen>0.0], breaks=lenbins ) ) )
## Now look by country
fp.c.counts <- with(fpblocks, table( cut( maplen[maplen>0.0], breaks=lenbins ), countrypair[maplen>0.0] ) )
fp.c.counts <- fp.c.counts[,match(names(npairs),dimnames(fp.c.counts)[[2]])] # make order agree with npairs
# fp.c.counts[is.na(fp.c.counts)] <- 0
true.c.counts <- with(obsblocks, table( cut( maplen[maplen>0.0], breaks=lenbins ), countrypair[maplen>0.0] ) )
true.c.counts <- true.c.counts[,match(names(npairs),dimnames(true.c.counts)[[2]])] # make order agree with npairs
# true.c.counts[is.na(true.c.counts)] <- 0
### only those we have at least 300 fpblocks for
manyblocks <- names(npairs)[ apply(fp.c.counts,2,sum,na.rm=TRUE)>300 ]
paircols <- rainbow(length(manyblocks))
names(paircols) <- manyblocks
## Absolute rate
# png(file="false-pos-by-country.png", width=7*144, height=5*144, res=144)
pdf(file="false-pos-by-country.pdf", width=8, height=5, pointsize=10)
plot(midbins, as.numeric(fp.counts/totalpairs)/diff(lenbins), type='l', xlim=c(0,10), lwd=2, ylim=c(0,.035), main="false positive rate")
abline(h=c(0,1))
invisible( sapply( manyblocks, function (k) lines( midbins, (fp.c.counts[,k]/npairs[k])/diff(lenbins), col=paircols[k] ) ) )
lines(midbins, exp(-7.108-1.879*midbins+3.705*sqrt(midbins)), lwd=2, lty=3)
legend("topright",legend=manyblocks, lty=1, col=paircols)
text( midbins[1], (fp.c.counts[1,manyblocks]/npairs[manyblocks])/diff(lenbins)[1], labels=manyblocks)
dev.off()
# Huh, look at PT-PT blocks compared to others
# pdf(file="example-false-pos.pdf", width=10,height=8,pointsize=10)
png(file="example-false-pos.png", width=10*144,height=8*144,res=144)
layout(1:3)
par(mar=c(2,4,2,2)+.1)
plotindivs(subset(fpblocks,(country1==country2)&(country2=="Portugal")),chroms=chromlist,main=paste("false positives, Portugal, n=",nsamples["Portugal"]))
theseones <- sample(subset(indivinfo,COUNTRY_SELF=="Switzerland")$SUBJID,nsamples["Portugal"])
plotindivs(subset(fpblocks,(id1%in%theseones)&(id2%in%theseones)),chroms=chromlist,main=paste("false positives, Switzerland, n=",nsamples["Portugal"]))
theseones <- sample(subset(indivinfo,COUNTRY_SELF=="Italy")$SUBJID,nsamples["Portugal"])
plotindivs(subset(fpblocks,(id1%in%theseones)&(id2%in%theseones)),chroms=chromlist,main=paste("false positives, Italy, n=",nsamples["Portugal"]))
dev.off()
# Yup; no doubt.
## Relative to observed rate
png(file="false-pos-relative.png", width=7*144, height=5*144, res=144)
plot(midbins, as.numeric(fp.counts/totalpairs)/as.numeric(true.counts/totalpairs), type='l', xlim=c(0,10), ylim=c(0,1))
abline(h=c(0,1), col='red')
abline(h=.1, lty=3, col='red')
dev.off()
# by country
pdf(file="false-pos-relative-by-country.pdf", width=7, height=5, pointsize=10)
par(mar=c(4,10,2,4)+.1)
matplot(midbins,(fp.c.counts/true.c.counts)[,manyblocks],type='l',col=rainbow(ncol(fp.c.counts)),yaxt='n', xlab="block length (cM)", ylab='', ylim=c(0,1), main="false positive rate relative to observed" )
axis(side=4);
axislab(side=2,at=(fp.c.counts/true.c.counts)[1,manyblocks], labels=manyblocks, las=2, cex.axis=8/12)
dev.off()
## Compared to observed rate
# png(file="false-pos-and-true-rate.png", width=7*144, height=5*144, res=144)
pdf(file="false-pos-and-true-rate.pdf", width=7, height=5, pointsize=10)
plot(midbins, sum(.chrlens)*as.numeric(true.counts/totalpairs)/(diff(lenbins)*sum(.chrlens[chromlist])), type='l', xlim=c(0,6), xlab="map length (cM)", ylab="rate per pair", main="False positive rates")
invisible( sapply( manyblocks, function (k) lines( midbins, sum(.chrlens)*(fp.c.counts[,k]/npairs[k])/(diff(lenbins)*sum(.chrlens[chromlist])), col=paircols[k] ) ) )
label.these <- (fp.c.counts[1,]/npairs)>.01
text( midbins[1], sum(.chrlens)*(fp.c.counts[1,]/npairs)[label.these]/(diff(lenbins)[1]*sum(.chrlens[chromlist])), labels=colnames(fp.c.counts)[label.these], cex=8/12, col=adjustcolor("black",.5))
lines(midbins, sum(.chrlens)*as.numeric(fp.counts/totalpairs)/(diff(lenbins)*sum(.chrlens[chromlist])), lty=2)
lines(midbins, sum(.chrlens)*exp(-13.704-2.095*midbins+4.381*sqrt(midbins)), lwd=2, lty=3)
legend("topright", legend=c("Observed mean IBD rate","Estimated mean false positive rate","Fitted mean fp rate","Estimated mean fp rate, by country-pair"), lty=c(1,2,3,1),col=c("black","black","black","red"))
# abline(h=c(0,1), col='red')
# abline(h=.1, lty=3, col='red')
dev.off()
# Look at short ones
makecp <- function(x,y) gsub(" ",".",paste(sort(c(x,y)),collapse="-"))
clist <- c("Italy","Spain","Portugal","Swiss.German","Swiss.French","United.Kingom","Germany","France","Ireland")
tmp <- sapply( clist, function (x) {
c( with( subset(fpblocks,countrypair==makecp(x,x)), sum( maplen>.4 & maplen<.5 ) ) / npairs[makecp(x,x)],
with( subset(obsblocks,countrypair==makecp(x,x)), sum( maplen>.4 & maplen<.5 ) ) / npairs[makecp(x,x)],
with( subset(fpblocks,countrypair==makecp(x,"Spain")), sum( maplen>.4 & maplen<.5 ) ) / npairs[makecp(x,"Spain")],
with( subset(obsblocks,countrypair==makecp(x,"Spain")), sum( maplen>.4 & maplen<.5 ) ) / npairs[makecp(x,"Spain")],
with( subset(fpblocks,countrypair==makecp(x,"Portugal")), sum( maplen>.4 & maplen<.5 ) ) / npairs[makecp(x,"Portugal")],
with( subset(obsblocks,countrypair==makecp(x,"Portugal")), sum( maplen>.4 & maplen<.5 ) ) / npairs[makecp(x,"Portugal")]
)
} )
barplot( data.frame( self=tmp[2,]-tmp[1,], ES=tmp[4,]-tmp[3,], PT=tmp[6,]-tmp[5,] ), beside=TRUE )
# Fit something to the mean block density per (total) maplen
lenbins <- quantile( obsblocks$maplen[obsblocks$maplen>0.5], seq(0,1,length.out=100), na.rm=TRUE )
midbins <- lenbins[-1]-diff(lenbins)/2
fp.counts <- with(fpblocks, table( cut( maplen[maplen>0.5], breaks=lenbins ) ) )
true.counts <- with(obsblocks, table( cut( maplen[maplen>0.5], breaks=lenbins ) ) )
plot(midbins, as.numeric(fp.counts/totalpairs)/as.numeric(true.counts/totalpairs), type='l', xlim=c(0,10), log='y')
fplm <- lm( log(as.numeric(fp.counts/totalpairs)/(diff(lenbins)*sum(.chrlens[chromlist]))) ~ midbins, subset=midbins>1 )
fplmsq <- lm( log(as.numeric(fp.counts/totalpairs)/(diff(lenbins)*sum(.chrlens[chromlist]))) ~ midbins + I(sqrt(midbins)), subset=midbins>1 & midbins<6)
# (Intercept) midbins I(sqrt(midbins))
# -7.108 -1.879 3.705
fplmsqsq <- lm( log(as.numeric(fp.counts/totalpairs)/(diff(lenbins)*sum(.chrlens[chromlist]))) ~ midbins + I(sqrt(midbins))+ I(midbins^(-1)), subset=midbins>1 )
# (Intercept) midbins I(sqrt(midbins)) I(midbins^2)
# -14.1369 -2.5989 5.2891 0.0402
par(mfrow=c(1,2))
plot(midbins, as.numeric(fp.counts/totalpairs)/(diff(lenbins)*sum(.chrlens[chromlist])), type='b', xlim=c(0,6), log='y')
lines( midbins, exp(predict( fplm, newdata=list(midbins=midbins) )), lty=3, col='red' )
lines( midbins, exp(predict( fplmsq, newdata=list(midbins=midbins) )), lty=2, col='green' )
lines( midbins, exp(predict( fplmsqsq, newdata=list(midbins=midbins) )), lty=4, col='purple' )
plot(midbins, as.numeric(fp.counts/totalpairs)/(diff(lenbins)*sum(.chrlens[chromlist])), type='b', xlim=c(0,300) )
lines( seq(0,300,length.out=100), exp(predict( fplm, newdata=list(midbins=seq(0,300,length.out=100)) )), lty=3, col='red' )
lines( seq(0,300,length.out=100), exp(predict( fplmsq, newdata=list(midbins=seq(0,300,length.out=100)) )), lty=2, col='green' )
lines( seq(0,300,length.out=100), exp(predict( fplmsqsq, newdata=list(midbins=seq(0,300,length.out=100)) )), lty=4, col='purple' )