‘vignettes/mesotrione_parent_2023.rnw’: Prebuilt vignette showing how covariate modelling can be done for all relevant parent degradation models.
‘inst/testdata/mesotrione_soil_efsa_2016}.xlsx’: Another example spreadsheets for use with ‘read_spreadsheet()’, featuring pH dependent degradation
R/illparms.R: Fix the detection of ill-defined slope or error model parameters for the case that the estimate is negative
‘R/{endpoints,parms,plot.mixed.mmkin,summary.saem.mmkin}.R’: Calculate parameters and endpoints and plot population curves for specific covariate values, or specific percentiles of covariate values used in saem fits.
Depend on current deSolve version with the possibility to avoid resolving symbols in a shared library (compiled models) over and over, thanks to Thomas Petzoldt.
‘inst/rmarkdown/templates/hierarchical_kinetics/skeleton/skeleton.Rmd’: Start a new cluster after creating a model stored in the user specified location, because otherwise symbols are not found by the worker processes.
‘tests/testthat/test_compiled_symbols.R’: Some new tests to control problems that may have been introduced by the possibility to use pre-resolved symbols.
‘R/mkinerrmin.R’: Fix typo in subset (use of = instead of ==), thanks to Sebastian Meyer for spotting this during his work on R 4.3.0.
‘inst/rmarkdown/templates/hierarchical_kinetics’: R markdown template to facilitate the application of hierarchical kinetic models.
‘inst/testdata/{cyantraniliprole_soil_efsa_2014,lambda-cyhalothrin_soil_efsa_2014}.xlsx’: Example spreadsheets for use with ‘read_spreadsheet()’.
‘R/mhmkin.R’: Allow an ‘illparms.mhmkin’ object or a list with suitable dimensions as value of the argument ‘no_random_effects’, making it possible to exclude random effects that were ill-defined in simpler variants of the set of degradation models. Remove the possibility to exclude random effects based on separate fits, as it did not work well.
‘R/summary.saem.mmkin.R’: List all initial parameter values in the summary, including random effects and error model parameters. Avoid redundant warnings that occurred in the calculation of correlations of the fixed effects in the case that the Fisher information matrix could not be inverted. List correlations of random effects if specified by the user in the covariance model.
‘R/parplot.R’: Possibility to select the top ‘llquant’ fraction of the fits for the parameter plots, and improved legend text.
‘R/illparms.R’: Also check if confidence intervals for slope parameters in covariate models include zero. Only implemented for fits obtained with the saemix backend.
‘R/parplot.R’: Make the function work also in the case that some of the multistart runs failed.
‘R/intervals.R’: Include correlations of random effects in the model in case there are any.
‘{data,R}/ds_mixed.rda’: Include the test data in the package instead of generating it in ‘tests/testthat/setup_script.R’. Refactor the generating code to make it consistent and update tests.
‘tests/testthat/setup_script.R’: Excluded another ill-defined random effect for the DFOP fit with ‘saem’, in an attempt to avoid a platform dependence that surfaced on Fedora systems on the CRAN check farm
‘tests/testthat/test_mixed.R’: Round parameters found by saemix to two significant digits before printing, to also help to avoid platform dependence of tests
‘R/saem.R’: Fix a bug that prevented that ‘error.ini’ is passed to ‘saemix_model’, and set default to c(1, 1) to avoid changing test results
‘R/parplot.R’: Show initial values for error model parameters
‘R/loglik.mkinfit.R’: Add ‘nobs’ attribute to the resulting ‘logLik’ object, in order to make test_AIC.R succeed on current R-devel
‘R/saem.R’: ‘logLik’, ‘update’ and ‘anova’ methods for ‘saem.mmkin’ objects.
‘R/saem.R’: Automatic estimation of start parameters for random effects for the case of mkin transformations, nicely improving convergence and reducing problems with iterative ODE solutions.
‘R/status.R’: New generic to show status information for fit array objects with methods for ‘mmkin’, ‘mhmkin’ and ‘multistart’ objects.
‘R/mhmkin.R’: New method for performing multiple hierarchical mkin fits in one function call, optionally in parallel.
‘R/mhmkin.R’: ‘anova.mhmkin’ for conveniently comparing the resulting fits.
‘R/illparms.R’: New generic to show ill-defined parameters with methods for ‘mkinfit’, ‘mmkin’, ‘saem.mmkin’ and ‘mhmkin’ objects.
‘R/multistart.R’: New method for testing multiple start parameters for hierarchical model fits, with function ‘llhist’ and new generic ‘parplot’ for diagnostics, and new generics ‘which.best’ and ‘best’ for extracting the fit with the highest likelihood
‘R/summary.mmkin.R’: Summary method for mmkin objects.
‘R/saem.R’: Implement and test saemix transformations for FOMC and HS. Also, error out if saemix transformations are requested but not supported.
‘R/read_spreadsheet.R’: Conveniently read in data from a spreadsheet file.
‘R/tex_listings.R’: Conveniently include summaries of fit objects in R markdown documents that are compiled to LaTeX.
’R/{mkinmod,mkinpredict}.R: Store DLL information in mkinmod objects and use that information in mkinpredict to avoid a performance regression brought by a bugfix in R 4.2.x. Thanks to Tomas Kalibera for his analysis of the problem on the r-package-devel list and his suggestion on how to fix it.
‘vignettes/FOCUS_L.rmd’: Remove an outdated note referring to a failure to calculate the covariance matrix for DFOP with the L2 dataset. Since 0.9.45.5 the covariance matrix is available
‘vignettes/web_only/benchmarks.rmd’: Add the first benchmark data using my laptop system, therefore add the CPU when showing the benchmark results.
‘dimethenamid_2018’: Update example code to use saemix
‘CAKE_export’: Check for validity of the map argument, updates
‘saem()’: Slightly improve speed in the case that no analytical solution for saemix is implemented, activate a test of the respective code
‘mean_degparms’: New argument ‘default_log_parms’ that makes it possible to supply a surrogate value (default) for log parameters that fail the t-test
‘plot.mixed.mmkin’: Pass the frame argument also to residual plots, take the ‘default_log_parms’ argument for ‘mean_degparms’ used for constructing approximate population curves, plot population curve last to avoid that it is covered by data
‘plot.mkinfit’: Respect argument ‘maxabs’ for residual plots, and make it possible to give ylim as a list, for row layouts
Reintroduce the interface to saemix version 3.0 (now on CRAN), in particular the generic function ‘saem’ with a generator ‘saem.mmkin’, currently using ‘saemix_model’ and ‘saemix_data’, summary and plot methods
‘mean_degparms’: New argument ‘test_log_parms’ that makes the function only consider log-transformed parameters where the untransformed parameters pass the t-test for a certain confidence level. This can be used to obtain more plausible starting parameters for the different mixed-effects model backends
‘plot.mixed.mmkin’: Gains arguments ‘test_log_parms’ and ‘conf.level’
‘vignettes/web_only/dimethenamid_2018.rmd’: Example evaluations of the dimethenamid data.
‘intervals’: Provide a method of this nlme function for ‘saem.mmkin’ objects.
All plotting functions setting graphical parameters: Use on.exit() for resetting graphical parameters
‘plot.mkinfit’: Use xlab and xlim for the residual plot if show_residuals is TRUE
‘mmkin’: Use cores = 1 per default on Windows to make it easier for first time users
‘confint.mmkin’, ‘nlme.mmkin’, ‘transform_odeparms’: Fix example code in dontrun sections that failed with current defaults
‘logLik.mkinfit’: Improve example code to avoid warnings and show convenient syntax
‘mkinresplot’: Re-add Katrin Lindenberger as coauthor who was accidentally removed long ago
Remove tests relying on non-convergence of the FOMC fit to the FOCUS A dataset as this is platform dependent (revealed by the new additional tests on CRAN, thanks!)
Increase test tolerance for some parameter comparisons that also proved to be platform dependent
‘mkinmod’ models gain arguments ‘name’ and ‘dll_dir’ which, in conjunction with a current version of the ‘inline’ package, make it possible to still use the DLL used for fast ODE solutions with ‘deSolve’ after saving and restoring the ‘mkinmod’ object.
‘mkindsg’ R6 class for groups of ‘mkinds’ datasets with metadata
‘f_norm_temp_focus’ generic function to normalise time intervals using the FOCUS method, with methods for numeric vectors and ‘mkindsg’ objects
‘D24_2014’ and ‘dimethenamid_2018’ datasets
‘focus_soil_moisture’ FOCUS default soil moisture data
‘update’ method for ‘mmkin’ objects
‘transform_odeparms’, ‘backtransform_odeparms’: Use logit transformation for solitary fractions like the g parameter of the DFOP model, or formation fractions for a pathway to only one target variable
‘plot.mmkin’: Add a ylab argument, making it possible to customize the y axis label of the panels on the left without affecting the residual plots. Reduce legend size and vertical distance between panels
‘plot.mkinfit’: Change default ylab from “Observed” to “Residue”. Pass xlab to residual plot if show_residuals is TRUE.
‘mixed.mmkin’ New container for mmkin objects for plotting with the ‘plot.mixed.mmkin’ method
‘plot.mixed.mmkin’ method used for ‘nlme.mmkin’ inheriting from ‘mixed.mmkin’ (currently virtual)
‘plot’, ‘summary’ and ‘print’ methods for ‘nlme.mmkin’ objects
‘parms’: Add a method for mmkin objects
‘mmkin’ and ‘confint(method = ’profile’): Use all cores detected by parallel::detectCores() per default
‘confint(method = ’profile’): Choose accuracy based on ‘rel_tol’ argument, relative to the bounds obtained by the quadratic approximation
‘mkinfit’: Make ‘use_of_ff’ = “max” also the default for models specified using short names like “SFO” or “FOMC”
‘mkinfit’: Run ‘stats::shapiro.test()’ on standardized residuals and warn if p < 0.05
‘mkinfit’: ‘error_model_algorithm’ = ‘d_3’ does not fail if direct fitting fails, but reports that the results for the threestep algorithm are returned
‘mmkin’: Do not fail any more if one of the fits fails, but assign the try-error to the respective position in the mmkin object
‘mkinfit’: Ignore components of state.ini that do not correspond to state variables in the model
‘endpoints’: Back-calculate DT50 value from DT90 also for the biphasic models DFOP, HS and SFORB
Increase tolerance for a platform specific test results on the Solaris test machine on CRAN
Updates and corrections (using the spelling package) to the documentation
Support SFORB with formation fractions
‘mkinmod’: Make ‘use_of_ff’ = “max” the default
Improve performance by a) avoiding expensive calls in the cost function like merge() and data.frame(), and b) by implementing analytical solutions for SFO-SFO and DFOP-SFO
‘nlme.mmkin’: An nlme method for mmkin row objects and an associated S3 class with print, plot, anova and endpoint methods
‘mean_degparms, nlme_data, nlme_function’: Three new functions to facilitate building nlme models from mmkin row objects
‘endpoints’: Don’t return the SFORB list component if it’s empty. This reduces distraction and complies with the documentation
Article in compiled models: Add some platform specific code and suppress warnings about zero values being removed from the FOCUS D dataset
‘plot.mmkin’: Add the argument ‘standardized’ to avoid warnings that occurred when it was passed as part of the additional arguments captured by the dots (…)
‘summary.mkinfit’: Add AIC, BIC and log likelihood to the summary
‘mkinmod’: Use pkgbuild::has_compiler instead of Sys.which(‘gcc’), as the latter will often fail even if Rtools are installed
‘mkinds’: Use roxygen for documenting fields and methods of this R6 class
‘aw’: Generic function for calculating Akaike weights, methods for mkinfit objects and mmkin columns
‘loftest’: Add a lack-of-fit test
‘plot_res’, ‘plot_sep’ and ‘mkinerrplot’: Add the possibility to show standardized residuals and make it the default for fits with error models other than ‘const’
‘lrtest.mkinfit’: Improve naming of the compared fits in the case of fixed parameters
‘confint.mkinfit’: Make the quadratic approximation the default, as the likelihood profiling takes a lot of time, especially if the fit has more than three parameters
Fix a bug introduced in 0.9.49.6 that occurred if the direct optimisation yielded a higher likelihood than the three-step optimisation in the d_3 algorithm, which caused the fitted parameters of the three-step optimisation to be returned instead of the parameters of the direct optimisation
Add a ‘nobs’ method for mkinfit objects, enabling the default ‘BIC’ method from the stats package. Also, add a ‘BIC’ method for mmkin column objects.
Implement a likelihood ratio test as a method for ‘lrtest’ from the lmtest package
Add an ‘update’ method for mkinfit objects which remembers fitted parameters if appropriate
Add a ‘residuals’ method for mkinfit objects that supports scaling based on the error model
Fix a bug in ‘mkinfit’ that prevented summaries of objects fitted with fixed parameters to be generated
Add ‘parms’ and ‘confint’ methods for mkinfit objects. Confidence intervals based on the quadratic approximation as in the summary, and based on the profile likelihood
Move long-running tests to tests/testthat/slow with a separate test log. They currently take around 7 minutes on my system
‘mkinfit’: Clean the code and return functions to calculate the log-likelihood and the sum of squared residuals
Vignette ‘twa.html’: Add the maximum time weighted average formulas for the hockey stick model
Support frameless plots (‘frame = FALSE’)
Support to suppress the chi2 error level (‘show_errmin = FALSE’) in ‘plot.mmkin’
Update README and the introductory vignette
Report ‘OLS’ as error_model_algorithm in the summary in the case that the default error_model (‘const’) is used
Support summarizing ‘mkinfit’ objects generated with versions < 0.9.49.5
Several algorithms for minimization of the negative log-likelihood for non-constant error models (two-component and variance by variable). In the case the error model is constant variance, least squares is used as this is more stable. The default algorithm ‘d_3’ tries direct minimization and a three-step procedure, and returns the model with the highest likelihood.
The argument ‘reweight.method’ to mkinfit and mmkin is now obsolete, use ‘error_model’ and ‘error_model_algorithm’ instead
Add a test that checks if we get the best known AIC for parent only fits to 12 test datasets. Add these test datasets for this purpose.
New function ‘mkinerrplot’. This function is also used for residual plots in ‘plot.mmkin’ if the argument ‘resplot = “errmod”’ is given, and in ‘plot.mkinfit’ if ‘show_errplot’ is set to TRUE.
Remove dependency on FME, only use nlminb for optimisation (‘Port’ algorithm). I cannot remember cases where one of the other optimisation algorithms was preferable, except that I sometime used Levenberg-Marquardt for speed in cases where I did not expect to get trapped in a local minimum.
Use the numDeriv package to calculate hessians. This results in slightly different confidence intervals, takes a bit longer, but is apparently more robust
Add a simple benchmark vignette to document the impact on performance.
The code for manual weighting was removed. This functionality might get added again at a later time. For the time being, please use an earlier version, e.g. 0.9.48.1 if you want to do manual weighting.
The fitting time reported in the summary now includes the time used for calculation of the hessians
Adapt tests
Fix an error in the FOCUS chi2 error level calculations that occurred if parameters were specified in parms.ini that were not in the model. A warning was already issued, but when fitting in parallel via mmkin this could go unnoticed.
Add example datasets obtained from risk assessment reports published by the European Food Safety Agency.
Add the function ‘logLik.mkinfit’ which makes it possible to calculate an AIC for mkinfit objects
Add the function ‘AIC.mmkin’ to make it easy to compare columns of mmkin objects
‘add_err’: Respect the argument giving the number of replicates in the synthetic dataset
‘max_twa_parent’: Support maximum time weighted average concentration calculations for the hockey stick (HS) model
‘mkinpredict’: Make the function generic and create a method for mkinfit objects
‘mkinfit’: Improve the correctness of the fitted two component error model by fitting the mean absolute deviance at each observation against the observed values, weighting with the current two-component error model
‘tests/testthat/test_irls.R’: Test if the components of the error model used to generate the data can be reproduced with moderate accuracy
Add the function ‘CAKE_export’ to facilitate cross-checking of results
Implement the logistic model (only tested for parent fits)
‘nafta’: Add evaluations according to the NAFTA guidance
Make the two-component error model stop in cases where it is inadequate to avoid nls crashes on windows
Move two vignettes to a location where they will not be built on CRAN (to avoid more NOTES from long execution times)
Exclude more example code from testing on CRAN to avoid NOTES from long execution times
‘mkinfit’: Improve fitting the error model for reweight.method = ‘tc’. Add ‘manual’ to possible arguments for ‘weight’
Test that FOCUS_2006_C can be evaluated with DFOP and reweight.method = ‘tc’
‘sigma_twocomp’: Rename ‘sigma_rl’ to ‘sigma_twocomp’ as the Rocke and Lorenzato model assumes lognormal distribution for large y. Correct references to the Rocke and Lorenzato model accordingly.
‘mkinfit’: Use 1.1 as starting value for N parameter of IORE models to obtain convergence in more difficult cases. Show parameter names when ‘trace_parms’ is ‘TRUE’.
Skip some tests on CRAN and winbuilder to avoid timeouts
‘test_data_from_UBA_2014’: Added this list of datasets containing experimental data used in the expertise from 2014
‘mkinfit’: Added the iterative reweighting method ‘tc’ using the two-component error model from Rocke and Lorenzato. NA values in the data are not returned any more.
‘mkinfit’: Work around a bug in the current FME version that prevented the convergence message to be returned in the case of non-convergence.
‘summary.mkinfit’: Improved output regarding weighting method. No predictions are returned for NA values in the model (see above).
‘summary.mkinfit’: Show versions of mkin and R used for fitting (not the ones used for the summary) if the fit was generated with mkin >= 0.9.47.1
README.md, vignettes/mkin.Rmd: URLs
were updated
synthetic_data_for_UBA: Add the code used to
generate the data in the interest of reproducibility
Converted the vignette FOCUS_Z from tex/pdf to markdown/html
DESCRIPTION: Add ORCID
plot.mkinfit: Fix scaling of residual plots for the
case of separate plots for each observed variable
plot.mkinfit: Use all data points of the fitted
curve for y axis scaling for the case of separate plots for each
observed variable
Documentation updates
test_FOMC_ill-defined.R as it is too platform
dependentRename twa to max_twa_parent to avoid
conflict with twa from my pfm package
Update URLs in documentation
Limit test code to one core to pass on windows
Switch from microbenchmark to
rbenchmark as the former is not supported on all
platforms
twa function, calculating maximum time weighted
average concentrations for the parent (SFO, FOMC and DFOP).plot.mkinfit and plot.mmkin: If the
plotting device is tikz, LaTeX markup is being used for the
chi2 error in the graphs.
Use pkgdown, the successor of
staticdocs for generating static HTML documentation.
Include example output and graphs also for dontrun
sections.
plot.mkinfit: Plotting does not fail any more when
the compiled model is not available, e.g. because it was removed from
the temporary directory. In this case, the uncompiled model is now used
for plotting
test_FOMC_ill-defined failed on several
architectures, so the test is now skippedThe title was changed to
Kinetic evaluations of chemical degradation data
plot.mkinfit: Add the possibility to show fits (and
residual plots if requested) separately for the observed
variables
plot.mkinfit: Add the possibility to show the chi2
error levels in the plot, similar to the way they are shown in
plot.mmkin
plot_sep: Add this function as a convenience wrapper
for plotting observed variables of mkinfit objects separately, with chi2
error values and residual plots.
Vignettes: The main vignette mkin was converted to R
markdown and updated. The other vignettes were also updated to show
current improved functionality.
The function add_err was added to the package,
making it easy to generate simulated data using an error model based on
the normal distribution
Remove an outdated reference to the inline package in the
compiled_models vignette
mkinfit: Do not error out in cases where the fit
converges, but the Jacobian for the untransformed model cost can not be
estimated. Give a warning instead and return NA for the t-test
results.
summary.mkinfit: Give a warning message when the
covariance matrix can not be obtained.
A test has been added to containing a corresponding edge case to check that the warnings are correctly issued and the fit does not terminate.
plot.mmkin: Round the chi2 error value to three
significant digits, instead of two decimal digits.
mkinfit: Return the err values used on
weighted fits as a column named err. Also include these
inverse weights when the column value in the observed data
is used, which is returned as observed in the data
component of the mkinfit object.
endpoints: When the name of a substance degrading to
a metabolite (e.g. a parent compound) used in the model formulation
ended in the letter f, some rate parameters could be listed
as formation fractions with mixed up names. These would also appear in
the summary.
mkinfit: Check for all observed variables when
checking if the user tried to fix formation fractions when fitting them
using ilr transformation.
plot.mmkin: Set the plot margins correctly, also in
the case of a single fit to be plotted, so the main title is placed in a
reasonable way.
plot.mkinfit: Correct default values for
col_obs, pch_obs and lty_obs for
the case that obs_vars is specified.
from_max_mean to mkinfit,
for fitting only the decline from the maximum observed value for models
with a single observed variableAdd plots to compiled_models vignette
Give an explanatory error message when mkinmod fails due to a missing definition of a target variable
print.mkinmod(): Improve formatting when printing
mkinmod model definitions
Add an R6 class mkinds representing datasets with a
printing method
Add a printing method for mkinmod objects
Make it possible to specify arbitrary strings as names for the
compounds in mkinmod, and show them in the plot
Use an index.r file to group help topics in static documentation
print.summary.mkinfit(): Avoid an error that occurred
when printing summaries generated with mkin versions before 0.9-36endpoints(): For DFOP and SFORB models, where
optimize() is used, make use of the fact that the DT50 must
be between DT50_k1 and DT50_k2 (DFOP) or DT50_b1 and DT50_b2 (SFORB), as
optimize() sometimes did not find the minimum. Likewise for
finding DT90 values. Also fit on the log scale to make the function more
efficient.DESCRIPTION, NAMESPACE,
R/*.R: Import (from) stats, graphics and methods packages,
and qualify some function calls for non-base packages installed with R
to avoid NOTES made by R CMD check –as-cran with upcoming R
versions.New function mmkin(): This function takes a
character vector of model shorthand names, or alternatively a list of
mkinmod models, as well as a list of dataset as main arguments. It
returns a matrix of mkinfit objects, with a row for each model and a
column for each dataset. A subsetting method with single brackets is
available. Fitting the models in parallel using the
parallel package is supported.
New function plot.mmkin(): Plots single-row or
single-column mmkin objects including residual
plots.
mkinparplot(): Fix the x axis scaling for rate
constants and formation fractions that got confused by the introduction
of the t-values of transformed parameters.vignettes/compiled_models.html: Show the performance
improvement factor actually obtained when building the vignette, as well
as mkin version, some system info and the CPU model used for building
the vignette.
GNUMakefile,vignettes/*: Clean up
vignette generation and include table of contents in HTML
vignettes.
mkinmod(): When generating the C code for the
derivatives, only declare the time variable when it is needed and remove
the ‘-W-no-unused-variable’ compiler flag as the C compiler used in the
CRAN checks on Solaris does not know it.summary.mkinfit(): A one-sided t-test for
significant difference of untransformed parameters from zero is now
always shown, based on the assumption of normal distribution for
estimators of all untransformed parameters. Use with caution, as this
assumption is unrealistic e.g. for rate constants in these nonlinear
kinetic models.
If a compiler (gcc) is installed, use a version of the differential equation model compiled from C code, which is a huge performance boost for models where only the deSolve method works.
mkinmod(): Create a list component $cf (of class
CFuncList) in the list returned by mkinmod, if a version can be compiled
from autogenerated C code (see above).
mkinfit(): Set the default
solution_type to deSolve when a compiled
version of the model is present, except when an analytical solution is
possible.
mkinparplot(): Avoid warnings that occurred when not
all confidence intervals were available in the summary of the
fit
print.summary.mkinfit(): Fix printing the summary
for the case that the number of iterations is not available
NAMESPACE: export S3 methods plot.mkinfit, summary.mkinfit and print.summary.mkinfit to satisfy R CMD check on R-devel
mkinparplot(): Avoid warning in R CMD check about
undeclared global variable Lower
mkinfit(): Report successful termination when quiet
= FALSE. This is helpful for more difficult problems fitted with
reweight.method = obs, as no progress is often indicated during the
reweighting.
A first test using results established in the expertise written for the German Federal Environmental Agency (UBA) was added.
Add synthetic datasets generated for expertise written for the German Federal Environmental Agency UBA
Add tests based on these datasets
Add the convenience function mkinsub() for creating
the lists used in mkinmod()
Add the possibility to fit indeterminate order rate equation (IORE) models using an analytical solution (parent only) or a numeric solution. Paths from IORE compounds to metabolites are supported when using formation fractions (use_of_ff = ‘max’). Note that the numerical solution (method.ode = ‘deSolve’) of the IORE differential equations sometimes fails due to numerical problems.
Switch to using the Port algorithm (using a model/trust region approach) per default. While needing more iterations than the Levenberg-Marquardt algorithm previously used per default, it is less sensitive to starting parameters.
The formatting of differential equations in the summary was further improved
Always include 0 on y axis when plotting during the fit
The initial value (state.ini) for the observed variable with the
highest observed residue is set to 100 in case it has no time zero
observation and state.ini = "auto"
A basic unit test for mkinerrmin() was
written
mkinfit(): The internally fitted parameter for
g was named g_ilr even when
transform_fractions=FALSE
mkinfit(): The initial value (state.ini) for the
parent compound was not set when the parent was not the (only) variable
with the highest value in the observed data.
mkinerrmin(): When checking for degrees of freedom
for metabolites, check if their time zero value is fixed instead of
checking if the observed value is zero. This ensures correct calculation
of degrees of freedom also in cases where the metabolite residue at time
zero is greater zero.
plot.mkinfit(): Avoid a warning message about only
using the first component of ylim that occurred when ylim was specified
explicitly
The formatting of differential equations in the summary was improved by wrapping overly long lines
The FOCUS_Z vignette was rebuilt with the above improvement and using a width of 70 to avoid output outside of the grey area
print.summary.mkinfit(): Avoid a warning that
occurred when gmkin showed summaries of initial fits without
iterations
mkinfit(): Avoid a warning that occurred when
summarising a fit that was performed with maxitmodFit = 0 as done in
gmkin for configuring new fits.
The number of degrees of freedom is difficult to define in the case of ilr transformation of formation fractions. Now for each source compartment the number of ilr parameters (=number of optimised parameters) is divided by the number of pathways to metabolites (=number of affected data series) which leads to fractional degrees of freedom in some cases.
The default for the initial value for the first state value is now taken from the mean of the observations at time zero, if available.
The kinetic model can alternatively be specified with a shorthand
name for parent only degradation models, e.g. SFO, or
DFOP.
Optimisation method, number of model evaluations and time elapsed during optimisation are given in the summary of mkinfit objects.
The maximum number of iterations in the optimisation algorithm
can be specified using the argument maxit.modFit to the
mkinfit function.
mkinfit gives a warning when the fit does not converge (does not apply to SANN method). This warning is included in the summary.
Avoid plotting an artificial 0 residual at time zero in
mkinresplot
In the determination of the degrees of freedom in
mkinerrmin, formation fractions were accounted for multiple
times in the case of parallel formation of metabolites. See the new
feature described above for the solution.
transform_rates=FALSE in mkinfit now
also works for FOMC and HS models.
Initial values for formation fractions were not set in all cases.
No warning was given when the fit did not converge when a method
other than the default Levenberg-Marquardt method Marq was
used.
Vignettes were rebuilt to reflect the changes in the summary method.
Algorithm Pseudo was excluded because it needs
user-defined parameter limits which are not supported.
Algorithm Newton was excluded because of its
different way to specify the maximum number of iterations and because it
does not appear to provide additional benefits.
mkinerrmin() used by the summary
function.mkinmod().The original and the transformed parameters now have different
names (e.g. k_parent and log_k_parent. They
also differ in how many they are when we have formation fractions but no
pathway to sink.
The order of some of the information blocks in
print.summary.mkinfit.R() has been ordered in a more
logical way.
The vignette FOCUS_Z has been simplified to use formation fractions with turning off the sink, and slightly amended to use the new versions of DT50 values calculated since mkin 0.9-29.
All vignettes have been rebuilt so they reflect all changes.
The ChangeLog was renamed to NEWS.md and the entries were
converted to markdown syntax compatible with the
tools::news() function built into R.
The test suite was overhauled. Tests of the DFOP and SFORB models with dataset FOCUS_2006_A were removed, as they were too much dependent on the optimisation algorithm and/or starting parameters, because the dataset is SFO (compare kinfit vignette).
Also, the Schaefer complex case can now be fitted using formation fractions, and with the ‘Port’ optimisation method we also fit A2 in the same way as published in the Piacenza paper.
Some more checks were introduced to mkinfit(),
leading to warnings or stopping execution if unsupported combinations of
methods and parameters are requested.
R/mkinresplot.R: Make it possible to specify
xlim
R/geometric_mean.R, man/geometric_mean.Rd: Add geometric mean function
R/endpoints.R, man/endpoints.Rd: Calculate additional (pseudo)-DT50 values for FOMC, DFOP, HS and SFORB. Avoid calculation of formation fractions from rate constants when they are directly fitted
Do not backtransform confidence intervals for formation fractions if more than one compound is formed, as such parameters only define the pathways as a set
Add historical remarks and some background to the main package vignette
Correct ‘isotropic’ into ‘isometric’ for the ilr transformation
Fork the GUI into a separate package gmkin
DESCRIPTION, NAMESPACE, TODO: Adapt and add copyright information
Remove files belonging to the GUI
Possibility to fit without parameter transformations, using bounds as implemented in FME
Add McCall 2,4,5-T dataset
Enable selection of observed variables in plotting
Add possibility to show residual plot in
plot.mkinfit
R/mkinparplot.R, man/mkinparplot.Rd: plot parameters with confidence intervals
Change vignette format from Sweave to knitr
Split examples vignette to FOCUS_L and FOCUS_Z
Remove warning about constant formation fractions in mkinmod as it was based on a misconception
Restrict the unit test with the Schaefer data to parent and primary metabolites as formation fraction and DT50 for A2 are highly correlated and passing the test is platform dependent. For example, the test fails in 1 out of 14 platforms on CRAN as of today.
Add Eurofins Regulatory AG copyright notices
Import FME and deSolve instead of depending on them to have clean startup
Add a starter function for the GUI: gmkin()
Change the format of the workspace files of gmkin so they can be distributed and documented in the package
Add gmkin workspace datasets FOCUS_2006_gmkin and FOCUS_2006_Z_gmkin
Bugfix re-enabling the fixing of any combination of initial values for state variables
Default values for kinetic rate constants are not all 0.1 any more but are “salted” with a small increment to avoid numeric artefacts with the eigenvalue based solutions
Backtransform fixed ODE parameters for the summary
Get rid of the optimisation step in mkinerrmin -
this was unnecessary. Thanks to KinGUII for the inspiration - actually
this is equation 6-2 in FOCUS kinetics p. 91 that I had overlooked
originally
Fix plot.mkinfit as it passed graphical arguments
like main to the solver
Do not use plot=TRUE in mkinfit()
example
The first successful fits in the not so simple GUI
Fix iteratively reweighted least squares for the case of many metabolites
Unify naming of initial values of state variables
Unify naming in dataframes of optimised and fixed parameters in the summary
Show the weighting method for residuals in the summary
Correct the output of the data in the case of manual weighting
Implement IRLS assuming different variances for observed variables
Do not use 0 values at time zero for chi2 error level calculations. This is the way it is done in KinGUII and it makes sense. It does impact the chi2 error levels in the output. Generally they seem to be lower for metabolites now, presumably because the mean of the observed values is higher
For a detailed list of changes to the mkin source please consult the commit history on http://github.com/jranke/mkin