Fit nonlinear mixed models with SAEM

Source: R/saemix.R

This function uses saemix::saemix() as a backend for fitting nonlinear mixed effects models created from mmkin row objects using the Stochastic Approximation Expectation Maximisation algorithm (SAEM).

saem(object, control, ...)

# S3 method for mmkin
saem(
  object,
  control = list(displayProgress = FALSE, print = FALSE, save = FALSE, save.graphs =
    FALSE),
  cores = 1,
  verbose = FALSE,
  suppressPlot = TRUE,
  ...
)

saemix_model(object, cores = 1, verbose = FALSE, ...)

saemix_data(object, verbose = FALSE, ...)

Arguments

object

An mmkin row object containing several fits of the same mkinmod model to different datasets
control

Passed to saemix::saemix
...

Further parameters passed to saemix::saemixData and saemix::saemixModel.
cores

The number of cores to be used for multicore processing using parallel::mclapply(). Using more than 1 core is experimental and may lead to uncontrolled forking, apparently depending on the BLAS version used.
verbose

Should we print information about created objects?
suppressPlot

Should we suppress any plotting that is done by the saemix function?

Value

An S3 object of class 'saem.mmkin', containing the fitted saemix::SaemixObject as a list component named 'so'.

An saemix::SaemixModel object.

An saemix::SaemixData object.

Details

An mmkin row object is essentially a list of mkinfit objects that have been obtained by fitting the same model to a list of datasets using mkinfit.

Starting values for the fixed effects (population mean parameters, argument psi0 of saemix::saemixModel() are the mean values of the parameters found using mmkin.

See also

summary.saem.mmkin

Examples

# \dontrun{
ds <- lapply(experimental_data_for_UBA_2019[6:10],
 function(x) subset(x$data[c("name", "time", "value")]))
names(ds) <- paste("Dataset", 6:10)
f_mmkin_parent_p0_fixed <- mmkin("FOMC", ds, cores = 1,
  state.ini = c(parent = 100), fixed_initials = "parent", quiet = TRUE)
f_saem_p0_fixed <- saem(f_mmkin_parent_p0_fixed)

#> Running main SAEM algorithm
#> [1] "Mon Nov  9 09:03:11 2020"
#> ....
#>     Minimisation finished
#> [1] "Mon Nov  9 09:03:13 2020"

f_mmkin_parent <- mmkin(c("SFO", "FOMC", "DFOP"), ds, quiet = TRUE)
f_saem_sfo <- saem(f_mmkin_parent["SFO", ])

#> Running main SAEM algorithm
#> [1] "Mon Nov  9 09:03:14 2020"
#> ....
#>     Minimisation finished
#> [1] "Mon Nov  9 09:03:16 2020"
f_saem_fomc <- saem(f_mmkin_parent["FOMC", ])

#> Running main SAEM algorithm
#> [1] "Mon Nov  9 09:03:16 2020"
#> ....
#>     Minimisation finished
#> [1] "Mon Nov  9 09:03:18 2020"
f_saem_dfop <- saem(f_mmkin_parent["DFOP", ])

#> Running main SAEM algorithm
#> [1] "Mon Nov  9 09:03:19 2020"
#> ....
#>     Minimisation finished
#> [1] "Mon Nov  9 09:03:22 2020"

# The returned saem.mmkin object contains an SaemixObject, therefore we can use
# functions from saemix
library(saemix)

#> Package saemix, version 3.1.9000
#>   please direct bugs, questions and feedback to emmanuelle.comets@inserm.fr
compare.saemix(list(f_saem_sfo$so, f_saem_fomc$so, f_saem_dfop$so))

#> Likelihoods computed by importance sampling 
#>        AIC      BIC
#> 1 624.2428 622.2900
#> 2 467.7644 465.0305
#> 3 491.3541 487.8391
plot(f_saem_fomc$so, plot.type = "convergence")

#> Plotting convergence plots
plot(f_saem_fomc$so, plot.type = "individual.fit")

#> Plotting individual fits
plot(f_saem_fomc$so, plot.type = "npde")

#> Simulating data using nsim = 1000 simulated datasets
#> Computing WRES and npde .
#> Plotting npde
#> ---------------------------------------------
#> Distribution of npde:
#>            mean= -0.01736   (SE= 0.098 )
#>        variance= 0.8562   (SE= 0.13 )
#>        skewness= 0.513 
#>        kurtosis= 1.202 
#> ---------------------------------------------
#> 
#> Statistical tests
#>   Wilcoxon signed rank test  : 0.652
#>   Fisher variance test       : 0.338
#>   SW test of normality       : 0.0757 .
#> Global adjusted p-value      : 0.227
#> ---
#> Signif. codes: '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 
#> ---------------------------------------------
plot(f_saem_fomc$so, plot.type = "vpc")

#> Performing simulations under the model.
#> Plotting VPC
#> Method used for VPC: binning by quantiles on X , dividing into the following intervals
#>   Interval  Centered.On
#> 1 (-1,3]      1.3      
#> 2 (3,8]       7.4      
#> 3 (8,14]     13.2      
#> 4 (14,21]    20.5      
#> 5 (21,37.7]  29.5      
#> 6 (37.7,60]  50.4      
#> 7 (60,90]    76.6      
#> 8 (90,120]  109.0      
#> 9 (120,180] 156.0      

f_mmkin_parent_tc <- update(f_mmkin_parent, error_model = "tc")
f_saem_fomc_tc <- saem(f_mmkin_parent_tc["FOMC", ])

#> Running main SAEM algorithm
#> [1] "Mon Nov  9 09:03:24 2020"
#> ....
#>     Minimisation finished
#> [1] "Mon Nov  9 09:03:29 2020"
compare.saemix(list(f_saem_fomc$so, f_saem_fomc_tc$so))

#> Likelihoods computed by importance sampling 
#>        AIC      BIC
#> 1 467.7644 465.0305
#> 2 469.4862 466.3617

sfo_sfo <- mkinmod(parent = mkinsub("SFO", "A1"),
  A1 = mkinsub("SFO"))

#> Successfully compiled differential equation model from auto-generated C code.
fomc_sfo <- mkinmod(parent = mkinsub("FOMC", "A1"),
  A1 = mkinsub("SFO"))

#> Successfully compiled differential equation model from auto-generated C code.
dfop_sfo <- mkinmod(parent = mkinsub("DFOP", "A1"),
  A1 = mkinsub("SFO"))

#> Successfully compiled differential equation model from auto-generated C code.
# The following fit uses analytical solutions for SFO-SFO and DFOP-SFO,
# and compiled ODEs for FOMC, both are fast
f_mmkin <- mmkin(list(
    "SFO-SFO" = sfo_sfo, "FOMC-SFO" = fomc_sfo, "DFOP-SFO" = dfop_sfo),
  ds, quiet = TRUE)
# These take about five seconds each on this system, as we use
# analytical solutions written for saemix. When using the analytical
# solutions written for mkin this took around four minutes
f_saem_sfo_sfo <- saem(f_mmkin["SFO-SFO", ])

#> Running main SAEM algorithm
#> [1] "Mon Nov  9 09:03:31 2020"
#> ....
#>     Minimisation finished
#> [1] "Mon Nov  9 09:03:36 2020"
f_saem_dfop_sfo <- saem(f_mmkin["DFOP-SFO", ])

#> Running main SAEM algorithm
#> [1] "Mon Nov  9 09:03:37 2020"
#> ....
#>     Minimisation finished
#> [1] "Mon Nov  9 09:03:46 2020"
summary(f_saem_dfop_sfo, data = FALSE)

#> saemix version used for fitting:      3.1.9000 
#> mkin version used for pre-fitting:  0.9.50.4 
#> R version used for fitting:         4.0.3 
#> Date of fit:     Mon Nov  9 09:03:47 2020 
#> Date of summary: Mon Nov  9 09:03:47 2020 
#> 
#> Equations:
#> d_parent/dt = - ((k1 * g * exp(-k1 * time) + k2 * (1 - g) * exp(-k2 *
#>            time)) / (g * exp(-k1 * time) + (1 - g) * exp(-k2 * time)))
#>            * parent
#> d_A1/dt = + f_parent_to_A1 * ((k1 * g * exp(-k1 * time) + k2 * (1 - g)
#>            * exp(-k2 * time)) / (g * exp(-k1 * time) + (1 - g) *
#>            exp(-k2 * time))) * parent - k_A1 * A1
#> 
#> Data:
#> 170 observations of 2 variable(s) grouped in 5 datasets
#> 
#> Model predictions using solution type analytical 
#> 
#> Fitted in 9.758 s using 300, 100 iterations
#> 
#> Variance model: Constant variance 
#> 
#> Mean of starting values for individual parameters:
#>        parent_0        log_k_A1 f_parent_qlogis          log_k1          log_k2 
#>      93.8101519      -9.7647455      -0.9711148      -1.8799371      -4.2708142 
#>        g_qlogis 
#>       0.1356441 
#> 
#> Fixed degradation parameter values:
#> None
#> 
#> Results:
#> 
#> Likelihood computed by importance sampling
#>        AIC      BIC    logLik
#>   841.3208 836.2435 -407.6604
#> 
#> Optimised, transformed parameters with symmetric confidence intervals:
#>                          est.     lower      upper
#> parent_0        93.7514328489 91.113651 96.3892150
#> log_k_A1        -6.1262333211 -8.432492 -3.8199749
#> f_parent_qlogis -0.9739851652 -1.371984 -0.5759863
#> log_k1          -2.4818388836 -3.746899 -1.2167788
#> log_k2          -3.6138616567 -5.294149 -1.9335743
#> g_qlogis        -0.0004613666 -1.063179  1.0622564
#> 
#> Correlation: 
#>                 prnt_0 lg__A1 f_prn_ log_k1 log_k2
#> log_k_A1        -0.013                            
#> f_parent_qlogis -0.025  0.050                     
#> log_k1           0.030  0.000 -0.005              
#> log_k2           0.013  0.005 -0.003  0.037       
#> g_qlogis        -0.068 -0.016  0.011 -0.181 -0.181
#> 
#> Random effects:
#>                         est.      lower     upper
#> SD.parent_0        2.7857084  0.7825105 4.7889063
#> SD.log_k_A1        2.1412505  0.4425207 3.8399803
#> SD.f_parent_qlogis 0.4463087  0.1609059 0.7317116
#> SD.log_k1          1.4097204  0.5240566 2.2953842
#> SD.log_k2          1.8739067  0.6979362 3.0498773
#> SD.g_qlogis        0.4559301 -0.8149852 1.7268453
#> 
#> Variance model:
#>         est.    lower    upper
#> a.1 1.882757 1.665681 2.099832
#> 
#> Backtransformed parameters with asymmetric confidence intervals:
#>                        est.        lower       upper
#> parent_0       93.751432849 9.111365e+01 96.38921497
#> k_A1            0.002184795 2.176784e-04  0.02192835
#> f_parent_to_A1  0.274086887 2.022995e-01  0.35985666
#> k1              0.083589373 2.359079e-02  0.29618269
#> k2              0.026947583 5.020885e-03  0.14463032
#> g               0.499884658 2.567024e-01  0.74312150
#> 
#> Resulting formation fractions:
#>                 ff
#> parent_A1   0.2741
#> parent_sink 0.7259
#> 
#> Estimated disappearance times:
#>          DT50    DT90 DT50back DT50_k1 DT50_k2
#> parent  13.91   60.89    18.33   8.292   25.72
#> A1     317.26 1053.91       NA      NA      NA

# Using a single core, the following takes about 6 minutes, using 10 cores
# it is slower instead of faster
#f_saem_fomc <- saem(f_mmkin["FOMC-SFO", ], cores = 1)
# }