-
Notifications
You must be signed in to change notification settings - Fork 4
/
StepwiseAnalysis.R
138 lines (93 loc) · 6.08 KB
/
StepwiseAnalysis.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
# Steve Horvath: Estimating DNAm age.
# This file assumes a data frame exists called dat1 whose rows correspond to CpGs
# and whose first column reports the CpG identifier
# and whose remaining columns corresponds to samples (e.g. Illumina arrays).
fastImputation=FALSE
#STEP 1: DEFINE QUALITY METRICS
meanMethBySample =as.numeric(apply(as.matrix(dat1[,-1]),2,mean,na.rm=TRUE))
minMethBySample =as.numeric(apply(as.matrix(dat1[,-1]),2,min,na.rm=TRUE))
maxMethBySample =as.numeric(apply(as.matrix(dat1[,-1]),2,max,na.rm=TRUE))
datMethUsed= t(dat1[,-1])
colnames(datMethUsed)=as.character(dat1[,1])
noMissingPerSample=apply(as.matrix(is.na(datMethUsed)),1,sum)
table(noMissingPerSample)
#STEP 2: Imputing
if (! fastImputation & nSamples>1 & max(noMissingPerSample,na.rm=TRUE)<3000 ){
# run the following code if there is at least one missing
if ( max(noMissingPerSample,na.rm=TRUE)>0 ){
dimnames1=dimnames(datMethUsed)
datMethUsed= data.frame(t(impute.knn(t(datMethUsed))$data))
dimnames(datMethUsed)=dimnames1
} # end of if
} # end of if (! fastImputation )
if ( max(noMissingPerSample,na.rm=TRUE)>=3000 ) fastImputation=TRUE
if ( fastImputation | nSamples==1 ){
noMissingPerSample=apply(as.matrix(is.na(datMethUsed)),1,sum)
table(noMissingPerSample)
if ( max(noMissingPerSample,na.rm=TRUE)>0 & max(noMissingPerSample,na.rm=TRUE) >= 3000 ) {normalizeData=FALSE}
# run the following code if there is at least one missing
if ( max(noMissingPerSample,na.rm=TRUE)>0 & max(noMissingPerSample,na.rm=TRUE) < 3000 ){
dimnames1=dimnames(datMethUsed)
for (i in which(noMissingPerSample>0) ){
selectMissing1=is.na(datMethUsed[i,])
datMethUsed[i,selectMissing1] = as.numeric(probeAnnotation21kdatMethUsed$goldstandard2[selectMissing1])
} # end of for loop
dimnames(datMethUsed)=dimnames1
} # end of if
} # end of if (! fastImputation )
# STEP 3: Data normalization (each sample requires about 8 seconds). It would be straightforward to parallelize this operation.
if (normalizeData ){
datMethUsedNormalized=BMIQcalibration(datM=datMethUsed,goldstandard.beta= probeAnnotation21kdatMethUsed$goldstandard2,plots=FALSE)
}
if (!normalizeData ){ datMethUsedNormalized=datMethUsed }
rm(datMethUsed); gc()
# Subset to common probes
common_probes <- intersect(as.character(datClock$CpGmarker[-1]), dimnames(datMethUsedNormalized)[[2]])
datClock <- datClock[datClock$CpGmarker %in% c("(Intercept)", common_probes), ]
datMethUsedNormalized <- datMethUsedNormalized[, colnames(datMethUsedNormalized) %in% common_probes]
#STEP 4: Predict age and create a data frame for the output (referred to as datout)
selectCpGsClock=is.element(dimnames(datMethUsedNormalized)[[2]], as.character(datClock$CpGmarker[-1]))
if ( sum( selectCpGsClock) < dim(datClock)[[1]]-1 ) {stop("The CpGs listed in column 1 of the input data did not contain the CpGs needed for calculating DNAm age. Make sure to input cg numbers such as cg00075967.")}
if ( sum( selectCpGsClock) > dim(datClock)[[1]]-1 ) {stop("ERROR: The CpGs listed in column 1 of the input data contain duplicate CpGs. Each row should report only one unique CpG marker (cg number).")}
if (nSamples>1 ) {
datMethClock0=data.frame(datMethUsedNormalized[,selectCpGsClock])
datMethClock= data.frame(datMethClock0[ as.character(datClock$CpGmarker[-1])])
dim(datMethClock)
predictedAge=as.numeric(anti.trafo(datClock$CoefficientTraining[1]+as.matrix(datMethClock)%*% as.numeric(datClock$CoefficientTraining[-1])))
} # end of if
if (nSamples==1 ) {
datMethUsedNormalized2=data.frame(rbind(datMethUsedNormalized,datMethUsedNormalized))
datMethClock0=data.frame(datMethUsedNormalized2[,selectCpGsClock])
datMethClock= data.frame(datMethClock0[ as.character(datClock$CpGmarker[-1])])
dim(datMethClock)
predictedAge=as.numeric(anti.trafo(datClock$CoefficientTraining[1]+as.matrix(datMethClock)%*% as.numeric(datClock$CoefficientTraining[-1])))
predictedAge=predictedAge[1]
} # end of if
# Let's add comments to the age prediction
Comment=ifelse ( predictedAge <0, "Negative DNAm age.", ifelse ( predictedAge >100, "Old DNAm age.", rep("",length(predictedAge))))
Comment[is.na(predictedAge)]="Age prediction was not possible. "
if ( sum( selectCpGsClock) < dim(datClock)[[1]]-1 ) {
Comment=rep("ERROR: The CpGs listed in column 1 of the input data did not contain the CpGs needed for calculating DNAm age. Make sure to input cg numbers such as cg00075967.",length(predictedAge) )}
if ( sum( selectCpGsClock) > dim(datClock)[[1]]-1 ) {
Comment=rep("ERROR: The CpGs listed in column 1 of the input data contain duplicate CpGs. Each row should report only one unique CpG marker (cg number).",length(predictedAge) )}
restSamples=-minMethBySample>0.05 | maxMethBySample>1.05;
restSamples[is.na(restSamples)]=FALSE
lab1="MAJOR WARNING: Probably you did not input beta values since either minMethBySample<-0.05 or maxMethBySample>1.05.";Comment[restSamples]= paste(Comment[restSamples],lab1)
restSamples= noMissingPerSample >0 & noMissingPerSample <=100;lab1="WARNING: Some beta values were missing, see noMissingPerSample."; Comment[restSamples]= paste(Comment[restSamples],lab1)
restSamples= noMissingPerSample >3000;lab1="MAJOR WARNING: More than 3k missing values!!"; Comment[restSamples]= paste(Comment[restSamples],lab1)
restSamples= noMissingPerSample >100 & noMissingPerSample <=3000 ;lab1="MAJOR WARNING: noMissingPerSample>100"
Comment[restSamples]= paste(Comment[restSamples],lab1)
restSamples=meanMethBySample>.35;
restSamples[is.na(restSamples)]=FALSE
lab1="Warning: meanMethBySample is >0.35";Comment[restSamples]= paste(Comment[restSamples],lab1)
restSamples=meanMethBySample<.25;
restSamples[is.na(restSamples)]=FALSE; lab1="Warning: meanMethBySample is <0.25"
Comment[restSamples]= paste(Comment[restSamples],lab1)
datout=data.frame(SampleID=colnames(dat1)[-1], DNAmAge=predictedAge, Comment, noMissingPerSample,meanMethBySample, minMethBySample, maxMethBySample)
if ( !is.null( meanXchromosome) ){
if ( length( meanXchromosome)==dim(datout)[[1]] ){
predictedGender=ifelse(meanXchromosome>.4,"female",
ifelse(meanXchromosome<.38,"male","Unsure"))
datout=data.frame(datout,predictedGender=predictedGender,meanXchromosome=meanXchromosome)
} # end of if
} # end of if