File: test-qc-outlier.R

package info (click to toggle)
r-bioc-scuttle 1.8.4%2Bdfsg-1
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 888 kB
  • sloc: cpp: 508; sh: 7; makefile: 2
file content (205 lines) | stat: -rw-r--r-- 7,797 bytes parent folder | download | duplicates (2)
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
# This tests the behaviour of the isOutlier() function.
# library(testthat); library(scuttle); source("test-qc-outlier.R")

set.seed(1000)
test_that("isOutlier works correctly with vanilla applications", {
    vals <- c(rnorm(10000), rnorm(100, sd=10))        
    out <- isOutlier(vals)
       
    # Checking that thresholds are correctly computed.
    MAD <- mad(vals)
    MED <- median(vals)
    lower <- MED - MAD * 3
    higher <- MED + MAD * 3
    expect_equal(attr(out, "thresholds"), c(lower=lower, higher=higher))
    expect_identical(as.logical(out), vals > higher | vals < lower)

    out.5 <- isOutlier(vals, nmads=5)
    expect_equal(attr(out.5, "thresholds"), c(lower=MED - MAD * 5, higher=MED + MAD * 5))

    # Consistent with just lower or higher.
    out.L <- isOutlier(vals, type="lower")
    out.H <- isOutlier(vals, type="higher")
    expect_identical(as.logical(out), out.L | out.H)
    
    expect_equal(attr(out.L, "thresholds"), c(lower=lower, higher=Inf))
    expect_equal(as.logical(out.L), vals < lower)

    expect_equal(attr(out.H, "thresholds"), c(lower=-Inf, higher=higher))
    expect_equal(as.logical(out.H), vals > higher)
})

set.seed(1001)
test_that("isOutlier responds to the minimum difference", {
    vals <- c(rnorm(10000), rnorm(100, sd=10))
    for (min.diff in c(1, 5, 10)) {
        out <- isOutlier(vals, min.diff=min.diff)
       
        # Checking that thresholds are correctly computed.
        relative.threshold <- max(min.diff, mad(vals) * 3)
        lower <- median(vals) - relative.threshold
        higher <- median(vals) + relative.threshold
        expect_equal(attr(out, "thresholds"), c(lower=lower, higher=higher))
        expect_identical(as.logical(out), vals > higher | vals < lower)
    }
})

set.seed(1002)
test_that("isOutlier responds to subsetting for threshold calculations", {
    vals <- c(rnorm(10000), rnorm(100, sd=10))
    chosen <- sample(length(vals), 1000)
    out <- isOutlier(vals, subset=chosen)   
    out2 <- isOutlier(vals[chosen])

    thresholds <- attr(out, "thresholds")
    expect_identical(thresholds, attr(out2, "thresholds"))
    expect_identical(out[chosen], out2)
    expect_identical(as.logical(out), vals < thresholds[1] | vals > thresholds[2])

    # NA inputs are automatically subsetted.
    vals.NA <- vals
    vals.NA[chosen] <- NA
    expect_warning(out.NA <- isOutlier(vals.NA), "missing values")
    ref.NA <- isOutlier(vals[-chosen])

    expect_identical(attr(out.NA, "thresholds"), attr(ref.NA, "thresholds"))
    expect_identical(out.NA[-chosen], ref.NA)
    expect_true(all(is.na(out.NA[chosen])))
})

set.seed(1003)
test_that("isOutlier responds to request for log-transformation", {
    vals <- c(1e-8, 100:200, 1000)
    by.log <- isOutlier(vals, log=TRUE)
    manual <- isOutlier(log(vals))
    expect_identical(as.logical(by.log), as.logical(manual))

    # Thresholds are un-adjusted for the log-transformation.
    thresh.log <- attr(by.log, "thresholds")
    thresh.man <- attr(manual, "thresholds")
    expect_equal(thresh.log, exp(thresh.man))
})

set.seed(1004)
test_that("isOutlier responds to batch specification", {
    vals <- c(rnorm(10000), rnorm(100, sd=10))
    batches <- sample(5, length(vals), replace=TRUE)
    out <- isOutlier(vals, batch=batches)

    for (b in unique(batches)) {
        chosen <- batches==b
        current <- isOutlier(vals[chosen])
        expect_identical(as.logical(current), as.logical(out[chosen]))
        expect_equal(attr(current, "thresholds"), attr(out, "thresholds")[,as.character(b)])
    }

    # Responds correctly when subset is also specified.
    sampled <- sample(length(vals), 1000)
    out <- isOutlier(vals, batch=batches, subset=sampled)

    for (b in unique(batches)) {
        chosen <- intersect(which(batches==b), sampled)
        current <- isOutlier(vals[chosen])
        expect_identical(as.logical(current), as.logical(out[chosen]))
        expect_equal(attr(current, "thresholds"), attr(out, "thresholds")[,as.character(b)])
    }
})

set.seed(10041)
test_that("isOutlier thresholds are computed correctly with batch specification", {
    vals <- rnorm(1000)
    batches <- gl(2, length(vals)/2)

    out <- isOutlier(vals, batch=batches)
    out1 <- isOutlier(vals[batches==1])
    out2 <- isOutlier(vals[batches==2])
    
    expect_equal(attr(out, "thresholds"), cbind(`1`=attr(out1, "thresholds"), `2`=attr(out2, "thresholds")))
    expect_identical(as.logical(out), as.logical(c(out1, out2)))

    # With log-transformation.
    vals <- rgamma(1000, 1, 1)
    out <- isOutlier(vals, batch=batches, log=TRUE)
    out1 <- isOutlier(vals[batches==1], log=TRUE)
    out2 <- isOutlier(vals[batches==2], log=TRUE)
    
    expect_equal(attr(out, "thresholds"), cbind(`1`=attr(out1, "thresholds"), `2`=attr(out2, "thresholds")))
    expect_identical(as.logical(out), as.logical(c(out1, out2)))
})

set.seed(10041)
test_that("isOutlier thresholds are computed correctly with sharingness", {
    vals <- rnorm(1000)
    batches <- gl(2, length(vals)/2)

    out <- isOutlier(vals, batch=batches, share.medians=TRUE, share.mads=TRUE)
    ref <- isOutlier(vals)
    expect_identical(as.logical(out), as.logical(ref))
    expect_identical(attr(out, "thresholds")[,1], attr(ref, "thresholds"))
    expect_identical(attr(out, "thresholds")[,2], attr(ref, "thresholds"))

    out <- isOutlier(vals, batch=batches, share.medians=TRUE)
    common.med <- median(vals)
    mad1 <- mad(vals[batches==1], center=common.med)
    mad2 <- mad(vals[batches==2], center=common.med)
    expect_equivalent(attr(out, "thresholds")[,1], common.med + c(-1, 1) * 3 * mad1)
    expect_equivalent(attr(out, "thresholds")[,2], common.med + c(-1, 1) * 3 * mad2)

    out <- isOutlier(vals, batch=batches, share.mads=TRUE)
    med1 <- median(vals[batches==1])
    med2 <- median(vals[batches==2])
    common.mad <- median(abs(vals - c(med1, med2)[batches])) * formals(mad)$constant
    expect_equivalent(attr(out, "thresholds")[,1], med1 + c(-1, 1) * 3 * common.mad)
    expect_equivalent(attr(out, "thresholds")[,2], med2 + c(-1, 1) * 3 * common.mad)
})

set.seed(10041)
test_that("isOutlier thresholds correctly recover missing batches", {
    vals <- rnorm(1000)
    batches <- gl(2, length(vals)/2)

    out <- isOutlier(vals, batch=batches, subset=batches==1)
    ref <- isOutlier(vals[batches==1])
    expect_identical(attr(out, "thresholds")[,1], attr(ref, "thresholds"))
    expect_identical(attr(out, "thresholds")[,2], attr(ref, "thresholds"))

    out <- isOutlier(vals, batch=batches, subset=batches==1, share.missing=FALSE)
    expect_equivalent(out[batches==1], ref)
    expect_true(all(is.na(out[batches==2])))
})    

set.seed(1005)
test_that("isOutlier handles silly inputs correctly", {
    out <- isOutlier(numeric(0))
    expect_identical(as.logical(out), logical(0))

    expect_error(out <- isOutlier(numeric(0), batch=1), "length of 'batch'")

    out <- isOutlier(1:10, subset=integer(0))
    expect_identical(as.logical(out), rep(NA, 10))
})

test_that("outlier.filter behaves correctly", {
    vals <- rnorm(1000)
    out <- isOutlier(vals)
    expect_s3_class(out, "outlier.filter")
    expect_true(!is.null(attr(out, "thresholds")))

    # Robust to filtering.
    expect_s3_class(out[1:10], "outlier.filter")
    expect_identical(attr(out[1:10], "thresholds"), attr(out, "thresholds"))

    # Works with c().
    combined <- c(out, logical(10))
    expect_null(attr(combined, "thresholds"))
    expect_type(combined, "logical")

    combined <- c(out, out)
    expect_null(attr(combined, "thresholds"))
    expect_type(combined, "logical")

    # We can use it for filtering.
    mock <- mockSCE()
    out <- isOutlier(colSums(counts(mock)), log=TRUE)
    expect_identical(mock[,out], mock[,as.logical(out)])
})