Data

Tools

Indexes

Applications

Experiments

Open Source Science

Recent Releases of nlmixr

nlmixr - v2.0.6

nlmixr 2.0.6

  • Fix focei subject initialization, see #566

- Rich Text Format
Published by mattfidler over 4 years ago

nlmixr - Version 2.0.5

  • Fix for nlmixrSim CMT to have a factor that matches the RxODE definition (issue #501)

  • Give instructions on how to reinstall nlmixr if it is linked to a different version of RxODE. (#555)

  • Now inform which parameters are near the boundary (#544)

  • The saem estimation routine will now increase the tolerance when ODE solving is difficult; This can be controlled with odeRecalcFactors and maxOdeRecalc. This is similar to the handling that focei already uses.

  • For focei family estimation methods:

    • If the inner problem couldn't solve the ODE using the forward sensitivities, try using numerical differences to approximate the derivatives needed for the focei problem. A warning will be issued when this occurs. This requires RxODE 1.1.0 that always generates the finite difference prediction model. If RxODE is an earlier version, only apply this when the finite differences are supplied to nlmixr. This occurs when there are ETAs on the dose based events like duration, lag time, bioavaibility etc.
    • If eta nudge is non-zero, when resetting an ETA estimate, try the zero estimate first, and then the nudged locations.
    • When there is an ODE system for an individual that cannot be optimized in the inner problem, adjust that individual's objective function by 100 points. This can be controlled by foceiControl(badSolveObjfAdj=100)
    • Theta reset now will now make sure the parameter is estimated and between the proper bounds before resetting.
    • $simInfo non longer tries to generate the covariance step, and will simply have a $simInfo$thetaMat entry of NULL if the covariance step was unsuccessful.
    • With vpc() if the cmt conversion isn't working correctly, fall back to compartment numbers.
    • Take out symbol stripping based on CRAN policies
    • Fall back gracefully when rbind doesn't work in parameter histories.
    • Correctly print out the number of compartments based on the new RxODE linCmt() that was updated to support solved systems in focei. (Reported by Bill Denney #537).
    • Use strict headers since Rcpp now is moving toward strict headers. Also changed all the Calloc to R_Calloc, Free to R_Free, and DOUBLE_EPS to DBL_EPSILON.

  • gnlmm no longer imports the data.frame to an RxODE event table. This should speed up the routine slightly and (more importantly) make it easier to specify time varying covariates.

nlmixr 2.0.4

  • Now can use the following for combinde error models: foceiControl(addProp=1) foceiControl(addProp=2) saemControl(addProp=1) saemControl(addProp=2)

  • Bug-fix for SAEM add+prop and other error models that are optimized with nelder mead simplex (#503)

  • Bug-fix for more complex SAEM models that were not parsing and running. (Issue

    502, #501)

  • Issue the "NaN in prediction" once per SAEM problem (#500)

- Rich Text Format
Published by mattfidler almost 5 years ago

nlmixr - v2.0.4

  • Now can use the following for combinde error models: foceiControl(addProp=1) foceiControl(addProp=2) saemControl(addProp=1) saemControl(addProp=2)

  • Bug-fix for SAEM add+prop and other error models that are optimized with nelder mead simplex (#503)

  • Bug-fix for more complex SAEM models that were not parsing and running. (Issue

    502, #501)

  • Issue the "NaN in prediction" once per SAEM problem (#500)

- Rich Text Format
Published by mattfidler over 5 years ago

nlmixr - v2.0.3

User interface changes

  • Detection of initial conditions was rewritten to enable additional features in the initial conditions (#322). The most important user-facing change is that now arbitrary R expressions can be used when setting initial conditions such as tvCL <- log(c(2,3,4)) (#253) instead of simply tvCL <- log(3)

  • The function as.nlmixrBounds() now supports adding the columns that are missing into the input data.frame.

  • omega definitions can be correlation matrices (#338)

  • Can specify keep= and drop= in the nlmixr function to keep and drop columns in nlmixr output. Can also specify control=list(keep=,drop=) or nlmixr(...,keep=,drop=) to keep/drop columns (#260)

focei changes:

  • Uses RxODE to re-arrange the problem so it does not include if/else in the model (ie. un-branched code). This allows sensitivities to be calculated in one pass saving time for multiple endpoint models and models with if/else in them.

    • linCmt() now uses solved systems instead of translating to ODEs.
    • Uses RxODE/stan's math headers to calculate the sensitivities of the super-positioned linCmt() solutions.
    • This uses the advan solutions and hence supports support time-varying covariates.
    • focei now supports censoring in the same way monolix does, with cens and limit columns
    • focei now allows etas on dose-related modeled events like alag, f, etc by finite difference sensitivities.
    • focei now supports 2 combined additive + proportional error models;
    • combined1: trans(y) = trans(f) + (a+b*f^c)*err
    • combined2: trans(y) = trans(f) + sqrt(a^2+b^2*f^(2c))*err
    • focei etaNudge parameters were changed to use quadrature points covering 95% percent of a standard normal.
    • With zero gradients, Gill differences are recomputed to try to find a non-zero gradient.
    • Now when running if a zero gradient is detected, reset the problem (theta reset) and re-estimated with outerOpt="bobyqa"
    • Now when running a model where the last objective function is not the minimum objective function, issue a warning and skip the covariance step. (See Issue #403)
    • focei proportional and power models are more tolerant of 0 predictions in your data

SAEM changes

  • saem fits now gracefully fall back to the focei likelihood when they support files are no longer on the loaded disk

  • saem phi pile is now saved in the RxODE::rxTempDir() which can be customized to allow the phi file to remain after R has exited

  • saem fits now can add in fo, foce and focei likelihood

  • saem fits now use liblsoda by default and are multi-threaded when running (controlled by RxODE)

  • saem now supports time-varying covariates (like clock-time)

  • saem now supports 2 combined additive + proportional error models:

    • combined1: trans(y) = trans(f) + (a+b*f^c)*err
    • combined2: trans(y) = trans(f) + sqrt(a^2+b^2*f^(2c))*err

  • saem proportional and power models are more tolerant of 0 predictions in your data

  • saem now supports censoring a similar way as monolix does, with cens and limit columns

  • The default of saem additive + proportional error has been switched to combined2, which was the focei default, but you can change this back with saemControl(addProp="combined2"). The table results will likely be different because in the last release the saem calculated combined1 and then used these coefficients in the combined2 focei problem.

nlme changes

  • nlme will now support 2 combined additive + proportional error models (if the patched version of nlme is used)

    • combined1: y = f + (a+b*f)*err
    • combined2: y = f + sqrt(a^2+b^2*f^2)*err
    • See https://github.com/nlmixrdevelopment/nlmixr/issues/428
    • Thanks to Johannes Ranke (@jranke) for the nlme patch and the catch

  • Can switch with nlmeControl(addProp="combined1") to use the combined1 type of error model

New Utilities

  • bootstrapFit now calculates the bootstrap confidence bands and (optionally) will compare with the theoretical chi-squared distribution to help assess their adequacy.

  • covarSearchAuto now allows automatic forward/backward covariate selection

General Changes

  • Added auto-completion of nlmixr object properties accessed by $. This works for major editors including Rstudio, ESS, and Base R itself.

  • Changed the way that Rstudio notebooks display nlmixr objects; It should be more legible in Rstudio.

  • Graphics have been revamped to show censoring (including adding ggplot stat/geom geom_cens) as well as use RxODE's ggplot theme (rxTheme()). Additionally time after dose is calculated as tad for all nlmixr models

  • Tables generation has been refactored; npde uses the arma and RxODE random number generators which may change results. Also the default of ties=TRUE has been changed to ties=FALSE. npde calculations have been threaded with OpenMP to speed up the calculation as well. This refactoring was required to have the dv imputation between cwres and npde use the same method. The npde option now calculates the decorrelated npd as well, (which is the recommended weighted residual; see Nguyen 2017)

Bug Fixes

  • Aligned saem and focei additive + proportional error models, so saem additive+proportional outputs will be different using the correct focei method

Note this includes all the RxODE changes including dropping python.

- Rich Text Format
Published by mattfidler over 5 years ago

nlmixr - CRAN v1.1.1-7

  • Update for solaris
  • Added clang USBAN fix

- Rich Text Format
Published by mattfidler over 6 years ago

nlmixr - CRAN release

  • Cran release to fix upcoming R's stringAsFactors change in R 4.0

- Rich Text Format
Published by mattfidler over 6 years ago

nlmixr - CRAN Release

This is the CRAN bug-fix release related to R version 4.0

It fixes the documentation and adds completion of the nlmixr object via $.

- Rich Text Format
Published by mattfidler over 6 years ago

nlmixr - nlmixr release v1.1.1-3

Compared to v1.1.1-2:

  • Some stability fixes for VPC and SAEM
  • Changed steady state handling
  • Memory allocation consolidated in preparation for threading and for spead increase
  • Moved nlmixr vignettes to pkgdown only for now.
  • Some augPred enhancements
  • Change correlation matrix printout handling
  • Fix mu-refereced covariates to be shown more consistently in the nlmixr(.) printout
  • Fixes related to new Stan headers

- Rich Text Format
Published by mattfidler over 6 years ago

nlmixr - Release Candidate for v1.1.1-3

This is the release candidate for v1.1.1-3

- Rich Text Format
Published by mattfidler almost 7 years ago

nlmixr - CRAN release

- Rich Text Format
Published by mattfidler almost 7 years ago

nlmixr - CRAN Release

Cran release; Per CRAN request will update.

- Rich Text Format
Published by mattfidler almost 7 years ago

nlmixr - CRAN 1.1.0-9

Installer changes

The nlmixr installer now has: - A readme describing: - How to link the installed version of nlmixr to Rstudio - How to make sure that only the .libPaths() of nlmixr are used in case there is a conflict - How to change your home-directory in case it is on a network drive (which can occasionally cause problems with packages and sometimes C++14 compile flags) - A shortcut to start shinyMixR directly, if you wish.

- Rich Text Format
Published by mattfidler about 7 years ago

nlmixr - CRAN v1.1.0-8

- Rich Text Format
Published by mattfidler about 7 years ago

nlmixr - v1.1.0-7

- Rich Text Format
Published by mattfidler over 7 years ago

nlmixr - v1.1.0-6

- Rich Text Format
Published by mattfidler over 7 years ago

nlmixr - v1.1.0-5

- Rich Text Format
Published by mattfidler over 7 years ago

nlmixr - v1.1.0-4

- Rich Text Format
Published by mattfidler over 7 years ago

nlmixr - v1.1.0-3

  • Bug fix for misplaced distributions.

- Rich Text Format
Published by mattfidler over 7 years ago

nlmixr - v1.1.0-2

  • Use augPred for default individual plots
  • Fix installer to work for more than just one developer (hopefully)

- Rich Text Format
Published by mattfidler over 7 years ago

nlmixr - v1.1.0-1

  • Bug fix for FOCEi w/ difficult initail values
  • Add theta-reset for FOCEi optimization with mu-referencing, producing better overall solutions for difficult FOCEi problems.

- Rich Text Format
Published by mattfidler over 7 years ago

nlmixr - v1.1.0-0

  • More stable SAEM/FOCEi algorithms
  • More robust estimation with FOCEi
  • Support of NONMEM-style datasets directly (except MDV and some other less used data items)
  • Align nlme/SAEM/FOCEi objective functions to be similar in value (and close or equal to NONMEM FOCEi objective function)
  • Align nlme/SAEM/FOCEi likelihood and likelihood measurments
  • Add model piping and a new model piping vignette
  • Add support for lag(), rate(), dur() and f() without between subject variability for all supported estimation methods

- Rich Text Format
Published by mattfidler over 7 years ago

nlmixr - FOCEi pre-release

  • FOCEi standard errors have been made more accurate

- Rich Text Format
Published by mattfidler almost 8 years ago

nlmixr - CRAN release

- Rich Text Format
Published by mattfidler almost 8 years ago

nlmixr - VPC fixes

- Rich Text Format
Published by mattfidler almost 8 years ago

nlmixr - Multiple Endpoint FO/FOi

A comprehensive collection of non-linear mixed effect (nlme) model algorithms:

  • First Order Estimation (FO) with FOCE post-hoc etas R ## eg nlmixr(modelfn, data, "fo"); # has FOCE post-hoc ETAs `
  • First Order Estimation (FO) with FOCEi post-hoc ETAs R ## eg nlmixr(modelfn, data, "foi"); # has FOCEi post-hoc ETAs `
  • First Order Conditional Estimate (FOCE) R ## eg nlmixr(modelfn, data, "foce"); `
  • First Order Conditional Estimate with interaction (FOCEi), R ## eg nlmixr(modelfn, data, "focei");
    • Adaptive Gaussian quadrature (with Laplacian approximation as a special case see glmm and glmm2),
    • Stochastic Approximation Estimation-Maximization (SAEM). R ## eg nlmixr(modelfn, data, "saem")
    • Traditional R nlme R ## eg nlmixr(modelfn, data, "nlme") # Other features
    • A minimalist, intuitive, expressive, and domain-specific nlme modeling language.
    • The capability of joint modeling of multiple endpoints.
    • A revamped SAEM engine with improved speed and stability.
    • The capability of out-of-box Visual Predictive Checks (VPC) after a model fit.
    • The capability of out-of-box sophisticated Clinical Trial Simulation (CTS) after a model fit.
    • The capability of an out-of-box comprehensive diagnostic kit with a direct hook to xpose after a model fit.
    • The capability of modeling “odd type” data, including binary data, count data, and bounded clinical endpoint (e.g., ADAS-cog has a range of 0 to 70).
    • Parallel computing ODE solving via the openmp package -- an industry's first among the current population PK/PD simulators to the best of our knowledge.
    • An intuitive, powerful, graphic user interface (GUI) based project manager in shinyMixR.

New in this release

This added a few things - fo and foi estimation routines - Multiple endpoint UI support: - SAEM bugfix

```R

Error in model block

cp1 is modeled by additive error for compartment depot

cp1 ~ add(add.err1) | depot cp2 ~ add(add.err2) | central ```

- Rich Text Format
Published by mattfidler almost 8 years ago

nlmixr - Add FO/FOi and multiple endpoint UI support.

A comprehensive collection of non-linear mixed effect (nlme) model algorithms:

  • First Order Estimation (FO) with FOCE post-hoc etas R ## eg nlmixr(modelfn, data, "fo"); # has FOCE post-hoc ETAs `
  • First Order Estimation (FO) with FOCEi post-hoc ETAs R ## eg nlmixr(modelfn, data, "foi"); # has FOCEi post-hoc ETAs `
  • First Order Conditional Estimate (FOCE) R ## eg nlmixr(modelfn, data, "foce"); `
  • First Order Conditional Estimate with interaction (FOCEi), R ## eg nlmixr(modelfn, data, "focei");
    • Adaptive Gaussian quadrature (with Laplacian approximation as a special case see glmm and glmm2),
    • Stochastic Approximation Estimation-Maximization (SAEM). R ## eg nlmixr(modelfn, data, "saem")
    • Traditional R nlme R ## eg nlmixr(modelfn, data, "nlme") # Other features
    • A minimalist, intuitive, expressive, and domain-specific nlme modeling language.
    • The capability of joint modeling of multiple endpoints. -c A revamped SAEM engine with improved speed and stability.
    • The capability of out-of-box Visual Predictive Checks (VPC) after a model fit.
    • The capability of out-of-box sophisticated Clinical Trial Simulation (CTS) after a model fit.
    • The capability of an out-of-box comprehensive diagnostic kit with a direct hook to xpose after a model fit.
    • The capability of modeling “odd type” data, including binary data, count data, and bounded clinical endpoint (e.g., ADAS-cog has a range of 0 to 70).
    • Parallel computing ODE solving via the openmp package -- an industry's first among the current population PK/PD simulators to the best of our knowledge.
    • An intuitive, powerful, graphic user interface (GUI) based project manager in shinyMixR.

New in this release

This added a few things - fo and foi estimation routines - Multiple endpoint UI support:

```R

Error in model block

cp1 is modeled by additive error for compartment depot

cp1 ~ add(add.err1) | depot cp2 ~ add(add.err2) | central ```

- Rich Text Format
Published by mattfidler almost 8 years ago

nlmixr - SAEM and FOCEi stabilization

A comprehensive collection of non-linear mixed effect (nlme) model algorithms:

  • First Order Conditional Estimate (FOCE) R ## eg nlmixr(modelfn, data, "foce"); `
  • First Order Conditional Estimate with interaction (FOCEi), R ## eg nlmixr(modelfn, data, "focei");
    • Adaptive Gaussian quadrature (with Laplacian approximation as a special case see glmm and glmm2),
    • Stochastic Approximation Estimation-Maximization (SAEM). R ## eg nlmixr(modelfn, data, "saem")
    • Traditional R nlme R ## eg nlmixr(modelfn, data, "nlme") # Other features
    • A minimalist, intuitive, expressive, and domain-specific nlme modeling language.
    • The capability of joint modeling of multiple endpoints.
    • A revamped SAEM engine with improved speed and stability.
    • The capability of out-of-box Visual Predictive Checks (VPC) after a model fit.
    • The capability of out-of-box sophisticated Clinical Trial Simulation (CTS) after a model fit.
    • The capability of an out-of-box comprehensive diagnostic kit with a direct hook to xpose after a model fit.
    • The capability of modeling “odd type” data, including binary data, count data, and bounded clinical endpoint (e.g., ADAS-cog has a range of 0 to 70).
    • Parallel computing ODE solving via the openmp package -- an industry's first among the current population PK/PD simulators to the best of our knowledge.
    • An intuitive, powerful, graphic user interface (GUI) based project manager in shinyMixR.

New in this release

This addresses a few stablitily fixes in the last release - SAEM FOCEi objective function was fixed - Standard Errors on FOCEi are more reasonable

- Rich Text Format
Published by mattfidler almost 8 years ago

nlmixr - v1.0.0-2

A comprehensive collection of non-linear mixed effect (nlme) model algorithms:

  • First Order Conditional Estimate (FOCE) R ## eg nlmixr(modelfn, data, "foce"); `
  • First Order Conditional Estimate with interaction (FOCEi), R ## eg nlmixr(modelfn, data, "focei");
    • Adaptive Gaussian quadrature (with Laplacian approximation as a special case see glmm and glmm2),
    • Stochastic Approximation Estimation-Maximization (SAEM). R ## eg nlmixr(modelfn, data, "saem")
    • Traditional R nlme R ## eg nlmixr(modelfn, data, "nlme") # Other features
    • A minimalist, intuitive, expressive, and domain-specific nlme modeling language.
    • The capability of joint modeling of multiple endpoints.
    • A revamped SAEM engine with improved speed and stability.
    • The capability of out-of-box Visual Predictive Checks (VPC) after a model fit.
    • The capability of out-of-box sophisticated Clinical Trial Simulation (CTS) after a model fit.
    • The capability of an out-of-box comprehensive diagnostic kit with a direct hook to xpose after a model fit.
    • The capability of modeling “odd type” data, including binary data, count data, and bounded clinical endpoint (e.g., ADAS-cog has a range of 0 to 70).
    • Parallel computing ODE solving via the openmp package -- an industry's first among the current population PK/PD simulators to the best of our knowledge.
    • An intuitive, powerful, graphic user interface (GUI) based project manager in shinyMixR.

- Rich Text Format
Published by mattfidler almost 8 years ago

nlmixr - v1.0.0-1

First major release

A comprehensive collection of non-linear mixed effect (nlme) model algorithms:

  • First Order Conditional Estimate (FOCE) R ## eg nlmixr(modelfn, data, "focei", control=foceiControl(interaction=FALSE)); `
  • First Order Conditional Estimate with interaction (FOCEi), R ## eg nlmixr(modelfn, data, "focei");
    • Adaptive Gaussian quadrature (with Laplacian approximation as a special case see glmm and glmm2),
    • Stochastic Approximation Estimation-Maximization (SAEM). R ## eg nlmixr(modelfn, data, "saem")
    • Traditional R nlme R ## eg nlmixr(modelfn, data, "nlme") # Other features
    • A minimalist, intuitive, expressive, and domain-specific nlme modeling language.
    • The capability of joint modeling of multiple endpoints.
    • A revamped SAEM engine with improved speed and stability.
    • The capability of out-of-box Visual Predictive Checks (VPC) after a model fit.
    • The capability of out-of-box sophisticated Clinical Trial Simulation (CTS) after a model fit.
    • The capability of an out-of-box comprehensive diagnostic kit with a direct hook to xpose after a model fit.
    • The capability of modeling “odd type” data, including binary data, count data, and bounded clinical endpoint (e.g., ADAS-cog has a range of 0 to 70).
    • Parallel computing ODE solving via the openmp package -- an industry's first among the current population PK/PD simulators to the best of our knowledge.
    • An intuitive, powerful, graphic user interface (GUI) based project manager in shinyMixR.

- Rich Text Format
Published by mattfidler almost 8 years ago

nlmixr - v0.9.1-3

CRAN release: - No longer depend on RxODE or nlme, rather import them instead.

- Rich Text Format
Published by mattfidler about 8 years ago

nlmixr - CRAN release

- Rich Text Format
Published by mattfidler about 8 years ago

nlmixr - PAGE 2018 Release (v9.0.1-0)

This is the PAGE 2018 release

- Rich Text Format
Published by mattfidler about 8 years ago

nlmixr - Minor Release to fix assertion errors

  • This release fixes the assertion errors displayed in Unix
  • This release alos allows parsing of meta information.

- Rich Text Format
Published by mattfidler over 8 years ago

nlmixr - nlmixr installer release

This release has some minor improvements; Additionally there are windows installers for this release.

The binaries include: - Python 3.6.2 - SymPy 1.1 - Rtools 3.4 - R 3.4.2 - R libraries - tidyverse - data.table - nlmixr and dependiencies (and most suggested packages) - xpose (from nlmixrdevelopment)

The i386 is for 32 bit platforms The x64 is for 64 bit platforms.

- Rich Text Format
Published by mattfidler over 8 years ago

nlmixr - First CRAN release

- Rich Text Format
Published by mattfidler over 8 years ago