aboutsummaryrefslogtreecommitdiff
path: root/tests/testthat/test_error_models.R
blob: c656f7d2abbbf13e914505f0b750818c5dc41837 (plain) (blame)
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
255
256
257
258
259
260
261
# Copyright (C) 2018,2019 Johannes Ranke
# Contact: jranke@uni-bremen.de

# This file is part of the R package mkin

# mkin is free software: you can redistribute it and/or modify it under the
# terms of the GNU General Public License as published by the Free Software
# Foundation, either version 3 of the License, or (at your option) any later
# version.

# This program is distributed in the hope that it will be useful, but WITHOUT
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
# FOR A PARTICULAR PURPOSE.  See the GNU General Public License for more
# details.

# You should have received a copy of the GNU General Public License along with
# this program. If not, see <http://www.gnu.org/licenses/>

context("Error model fitting")

m_synth_SFO_lin <- mkinmod(parent = mkinsub("SFO", "M1"),
                           M1 = mkinsub("SFO", "M2"),
                           M2 = mkinsub("SFO"),
                           use_of_ff = "max", quiet = TRUE)

m_synth_DFOP_par <- mkinmod(parent = mkinsub("DFOP", c("M1", "M2")),
                           M1 = mkinsub("SFO"),
                           M2 = mkinsub("SFO"),
                           use_of_ff = "max", quiet = TRUE)

SFO_lin_a <- synthetic_data_for_UBA_2014[[1]]$data

DFOP_par_c <- synthetic_data_for_UBA_2014[[12]]$data

test_that("Error model 'const' works", {
  skip_on_cran()
  fit_const_1 <- mkinfit(m_synth_SFO_lin, SFO_lin_a, error_model = "const", quiet = TRUE)
  bpar_1 <- summary(fit_const_1)$bpar[, c("Estimate", "Lower", "Upper")]
  # The reference used here is mkin 0.9.48.1
  bpar_1_mkin_0.9 <-   read.table(text =
"parent_0       102.0000 98.6000 106.0000
k_parent         0.7390  0.6770   0.8070
k_M1             0.2990  0.2560   0.3490
k_M2             0.0202  0.0176   0.0233
f_parent_to_M1   0.7690  0.6640   0.8480
f_M1_to_M2       0.7230  0.6030   0.8180",
col.names = c("parameter", "estimate", "lower", "upper"))

  expect_equivalent(signif(bpar_1[1:6, "Estimate"], 3), bpar_1_mkin_0.9$estimate)
  # Relative difference of lower bound of confidence is < 0.02
  rel_diff <- function(v1, v2) {
    (v1 - v2)/v2
  }
  expect_equivalent(rel_diff(bpar_1[1:6, "Lower"],
                             bpar_1_mkin_0.9$lower),
                    rep(0, 6), tolerance = 0.02)
})

test_that("Error model 'obs' works", {
  skip_on_cran()
  fit_obs_1 <- mkinfit(m_synth_SFO_lin, SFO_lin_a, error_model = "obs", quiet = TRUE)
  parms_2 <- round(fit_obs_1$bparms.optim, c(1, 4, 4, 4, 4, 4))
  expect_equivalent(parms_2, c(102.1, 0.7389, 0.2982, 0.0203, 0.7677, 0.7246))
})

test_that("Error model 'tc' works", {
  skip_on_cran()
  fit_tc_1 <- mkinfit(m_synth_SFO_lin, SFO_lin_a, error_model = "tc", quiet = TRUE)
  parms_3 <- round(fit_tc_1$bparms.optim, c(1, 4, 4, 4, 4, 4))
  expect_equivalent(parms_3, c(102.1, 0.7393, 0.2992, 0.0202, 0.7687, 0.7229))
})

test_that("Reweighting method 'tc' produces reasonable variance estimates", {

  # Check if we can approximately obtain the parameters and the error model
  # components that were used in the data generation

  # Parent only
  DFOP <- mkinmod(parent = mkinsub("DFOP"))
  sampling_times = c(0, 1, 3, 7, 14, 28, 60, 90, 120)
  parms_DFOP <- c(k1 = 0.2, k2 = 0.02, g = 0.5)
  parms_DFOP_optim <- c(parent_0 = 100, parms_DFOP)

  d_DFOP <- mkinpredict(DFOP,
     parms_DFOP, c(parent = 100),
     sampling_times)
  d_2_10 <- add_err(d_DFOP,
    sdfunc = function(x) sigma_twocomp(x, 0.5, 0.07),
    n = 10, reps = 2, digits = 5, LOD = -Inf, seed = 123456)
  d_100_1 <- add_err(d_DFOP,
    sdfunc = function(x) sigma_twocomp(x, 0.5, 0.07),
    n = 1, reps = 100, digits = 5, LOD = -Inf, seed = 123456)

  # Per default (on my box) use all cores minus one
  n_cores <- parallel::detectCores() - 1

  # We are only allowed one core on travis
  if (Sys.getenv("TRAVIS") != "") n_cores = 1

  # Also on Windows we would need to make a cluster first,
  # and I do not know how this would work on winbuilder or CRAN, so
  if (Sys.info()["sysname"] == "Windows") n_cores = 1

  # Unweighted fits
  f_2_10 <- mmkin("DFOP", d_2_10, error_model = "const", quiet = TRUE,
    cores = n_cores)
  parms_2_10 <- apply(sapply(f_2_10, function(x) x$bparms.optim), 1, mean)
  parm_errors_2_10 <- (parms_2_10 - parms_DFOP_optim) / parms_DFOP_optim
  expect_true(all(abs(parm_errors_2_10) < 0.12))

  f_2_10_tc <- mmkin("DFOP", d_2_10, error_model = "tc", quiet = TRUE,
    cores = n_cores)
  parms_2_10_tc <- apply(sapply(f_2_10_tc, function(x) x$bparms.optim), 1, mean)
  parm_errors_2_10_tc <- (parms_2_10_tc - parms_DFOP_optim) / parms_DFOP_optim
  expect_true(all(abs(parm_errors_2_10_tc) < 0.05))

  tcf_2_10_tc <- apply(sapply(f_2_10_tc, function(x) x$errparms), 1, mean, na.rm = TRUE)

  tcf_2_10_error_model_errors <- (tcf_2_10_tc - c(0.5, 0.07)) / c(0.5, 0.07)
  expect_true(all(abs(tcf_2_10_error_model_errors) < 0.2))

  # When we have 100 replicates in the synthetic data, we can roundtrip
  # the parameters with < 2% precision
  f_tc_100_1 <- mkinfit(DFOP, d_100_1[[1]], error_model = "tc", quiet = TRUE)
  parm_errors_100_1 <- (f_tc_100_1$bparms.optim - parms_DFOP_optim) / parms_DFOP_optim
  expect_true(all(abs(parm_errors_100_1) < 0.02))

  tcf_100_1_error_model_errors <- (f_tc_100_1$errparms - c(0.5, 0.07)) /
    c(0.5, 0.07)
  # When maximising the likelihood directly (not using IRLS), we get
  # a precision of < 2% for the error model componentes as well
  expect_true(all(abs(tcf_100_1_error_model_errors) < 0.02))

  # Parent and two metabolites
  m_synth_DFOP_lin <- mkinmod(parent = list(type = "DFOP", to = "M1"),
                             M1 = list(type = "SFO", to = "M2"),
                             M2 = list(type = "SFO"), use_of_ff = "max",
                             quiet = TRUE)
  sampling_times = c(0, 1, 3, 7, 14, 28, 60, 90, 120)
  parms_DFOP_lin <- c(k1 = 0.2, k2 = 0.02, g = 0.5,
     f_parent_to_M1 = 0.5, k_M1 = 0.3,
     f_M1_to_M2 = 0.7, k_M2 = 0.02)
  d_synth_DFOP_lin <- mkinpredict(m_synth_DFOP_lin,
     parms_DFOP_lin,
     c(parent = 100, M1 = 0, M2 = 0),
     sampling_times)
  parms_DFOP_lin_optim = c(parent_0 = 100, parms_DFOP_lin)

  d_met_2_15 <- add_err(d_synth_DFOP_lin,
    sdfunc = function(x) sigma_twocomp(x, 0.5, 0.07),
    n = 15, reps = 100, digits = 5, LOD = 0.01, seed = 123456)

  # For a single fit, we get a relative error of less than 10%  in the error
  # model components
  f_met_2_tc_e4 <- mkinfit(m_synth_DFOP_lin, d_met_2_15[[1]], quiet = TRUE,
                          error_model = "tc")
  parm_errors_met_2_tc_e4 <- (f_met_2_tc_e4$errparms - c(0.5, 0.07)) / c(0.5, 0.07)
  expect_true(all(abs(parm_errors_met_2_tc_e4) < 0.1))

  # Doing more takes a lot of computing power
  skip_on_travis()
  skip_on_cran()
  f_met_2_15_tc_e4 <- mmkin(list(m_synth_DFOP_lin), d_met_2_15, quiet = TRUE,
                            error_model = "tc", cores = n_cores)

  parms_met_2_15_tc_e4 <- apply(sapply(f_met_2_15_tc_e4, function(x) x$bparms.optim), 1, mean)
  parm_errors_met_2_15_tc_e4 <- (parms_met_2_15_tc_e4[names(parms_DFOP_lin_optim)] -
                                 parms_DFOP_lin_optim) / parms_DFOP_lin_optim
  expect_true(all(abs(parm_errors_met_2_15_tc_e4) < 0.015))

  tcf_met_2_15_tc <- apply(sapply(f_met_2_15_tc_e4, function(x) x$errparms), 1, mean, na.rm = TRUE)

  tcf_met_2_15_tc_error_model_errors <- (tcf_met_2_15_tc - c(0.5, 0.07)) /
    c(0.5, 0.07)

  # Here we get a precision < 10% for retrieving the original error model components
  # from 15 datasets
  expect_true(all(abs(tcf_met_2_15_tc_error_model_errors) < 0.10))
})

test_that("The two-component error model finds the best known AIC values for parent models", {
  skip_on_cran()
  library(parallel)
  source("~/git/mkin/R/mkinfit.R")
  source("~/git/mkin/R/mmkin.R")
  f_9 <- mkinfit("SFO", experimental_data_for_UBA_2019[[9]]$data)
  f_9 <- mkinfit("SFO", experimental_data_for_UBA_2019[[9]]$data,
                 error_model = "tc", error_model_algorithm = "direct")
  f_9 <- mkinfit("SFO", experimental_data_for_UBA_2019[[9]]$data,
                 error_model = "tc", error_model_algorithm = "twostep")
  f_9 <- mkinfit("SFO", experimental_data_for_UBA_2019[[9]]$data,
                 error_model = "tc", error_model_algorithm = "threestep")
  f_9 <- mkinfit("SFO", experimental_data_for_UBA_2019[[9]]$data,
                 error_model = "tc", error_model_algorithm = "fourstep")
  f_9 <- mkinfit("SFO", experimental_data_for_UBA_2019[[9]]$data,
                 error_model = "tc", error_model_algorithm = "IRLS")
  f_9 <- mkinfit("SFO", experimental_data_for_UBA_2019[[9]]$data,
                 error_model = "tc", error_model_algorithm = "d_3")
  AIC(f_9)
  f_10 <- mkinfit("DFOP", experimental_data_for_UBA_2019[[10]]$data,
                 error_model = "tc", error_model_algorithm = "IRLS")
  f_tc_exp_direct <- mmkin(c("SFO", "DFOP", "HS"),
    lapply(experimental_data_for_UBA_2019, function(x) x$data),
    error_model = "tc",
    error_model_algorithm = "direct",
    quiet = TRUE)
  f_tc_exp_twostep <- mmkin(c("SFO", "DFOP", "HS"),
    lapply(experimental_data_for_UBA_2019, function(x) x$data),
    error_model = "tc",
    error_model_algorithm = "twostep",
    quiet = TRUE)
  f_tc_exp_threestep <- mmkin(c("SFO", "DFOP", "HS"),
    lapply(experimental_data_for_UBA_2019, function(x) x$data),
    error_model = "tc",
    error_model_algorithm = "threestep",
    quiet = TRUE)
  f_tc_exp_fourstep <- mmkin(c("SFO", "DFOP", "HS"),
    lapply(experimental_data_for_UBA_2019, function(x) x$data),
    error_model = "tc",
    error_model_algorithm = "fourstep",
    quiet = TRUE)
  f_tc_exp_IRLS <- mmkin(c("SFO", "DFOP", "HS"),
    lapply(experimental_data_for_UBA_2019, function(x) x$data),
    error_model = "tc",
    error_model_algorithm = "IRLS",
    quiet = TRUE)

  AIC_exp_direct <- lapply(f_tc_exp_direct, AIC)
  AIC_exp_direct <- lapply(AIC_exp_direct, round, 1)
  dim(AIC_exp_direct) <- dim(f_tc_exp_direct)
  dimnames(AIC_exp_direct) <- dimnames(f_tc_exp_direct)

  AIC_exp_twostep <- lapply(f_tc_exp_twostep, AIC)
  AIC_exp_twostep <- lapply(AIC_exp_twostep, round, 1)
  dim(AIC_exp_twostep) <- dim(f_tc_exp_twostep)
  dimnames(AIC_exp_twostep) <- dimnames(f_tc_exp_twostep)

  AIC_exp_threestep <- lapply(f_tc_exp_threestep, AIC)
  AIC_exp_threestep <- lapply(AIC_exp_threestep, round, 1)
  dim(AIC_exp_threestep) <- dim(f_tc_exp_threestep)
  dimnames(AIC_exp_threestep) <- dimnames(f_tc_exp_threestep)

  AIC_exp_fourstep <- lapply(f_tc_exp_fourstep, AIC)
  AIC_exp_fourstep <- lapply(AIC_exp_fourstep, round, 1)
  dim(AIC_exp_fourstep) <- dim(f_tc_exp_fourstep)
  dimnames(AIC_exp_fourstep) <- dimnames(f_tc_exp_fourstep)

  AIC_exp_IRLS <- lapply(f_tc_exp_IRLS, AIC)
  AIC_exp_IRLS <- lapply(AIC_exp_IRLS, round, 1)
  dim(AIC_exp_IRLS) <- dim(f_tc_exp_IRLS)
  dimnames(AIC_exp_IRLS) <- dimnames(f_tc_exp_IRLS)

  AIC_exp <- lapply(f_tc_exp, AIC)
  dim(AIC_exp) <- dim(f_tc_exp)
  dimnames(AIC_exp) <- dimnames(f_tc_exp)
  unlist(AIC_exp["SFO", c(9, 11, 12)])
  expect_equivalent(round(unlist(AIC_exp["SFO", c(9, 11, 12)]), 1),
                    c(134.9, 125.5, 82.0))
})


Contact - Imprint