한국어 설명은 README.ko.md 에 있습니다.
The R code of the textbook 『약동학 with R』 (Pharmacokinetics with R) by Kyun-Seop Bae.
The code printed in the book is the file in this repository, and the console output and the figures printed in the book were produced by running it. So the code in the book cannot drift away from the code that actually ran. Changing a value and running it again is the way this book asks to be studied.
The book is written in Korean, and so are the comments in nearly all of the
code files. Everything else is not language-bound: the file names, the chapter
numbering, the console output and the figures are the same in any language, and
the TDM-Vanco web app below has an English interface. An English edition of
the book is in preparation, and its code is already here: En/ holds the same
snippets with the comments, labels and messages in English, together with their
own frozen output and figures (see below).
| Folder | Contents |
|---|---|
R/snippets/ |
201 snippets, one file per name. The NN of chNN-name.R is the chapter number. The order they appear in the book is the order in R/build.R, not alphabetical |
output/ |
the console output of each snippet, 200 files. ch11-ctl is a NONMEM control stream shown as an example and is not run |
figures/ |
the 82 figures the snippets make |
R/build.R |
remakes all of it. The session boundaries between chapters are in its comments |
R/_freeze.R |
the tool that runs one snippet and freezes its output and figure |
En/ |
the same for the English edition: translated snippets, their output and figures, and En/build.R (below) |
ACRE/ |
the syntax-highlighting definitions (Syntax/) of the ACRE editor that §1.11 introduces, and its README |
TDM-Vanco/ |
the vancomycin TDM Shiny app of §13.7 (below) |
Rscript R/build.R # the Korean edition's output/ and figures/
Rscript En/build.R # the English edition's En/output/ and En/figures/Run from the repository root, R/build.R remakes output/ and figures/ from
scratch. It takes 12–15 minutes, most of it the FOCE-I estimation of Chapter 11
and the randomization test and bootstrap of Chapter 12. To look at one snippet,
open that file and just run it.
The packages needed are these. A star marks one the author wrote.
| Package | Chapters | What it does |
|---|---|---|
NonCompart* |
5, 9, 10 | non-compartmental analysis (NCA) |
wnl* |
4, 7, 8, 16, 17 | nonlinear regression and error models |
nmw* |
11, 13 | NONMEM's estimation steps, reproduced in R |
BE* |
14 | bioequivalence analysis |
sasLM* |
14 | linear models in the form of SAS PROC GLM |
LBI* |
3 | likelihood intervals |
mathr* |
2 | numerical tools (machine epsilon, quadrature, numerical differentiation) |
deSolve |
2, 5, 7, 8, 10, 13, 17 | numerical solution of differential equations |
PowerTOST |
14 | power and sample size for equivalence trials |
nlme |
5, 14 | example data and mixed-effects models |
install.packages(c("NonCompart", "wnl", "nmw", "BE", "sasLM", "LBI", "mathr",
"deSolve", "PowerTOST", "nlme"))Some chapters share a session. Most snippets run in a fresh R session per chapter, but at three places a chapter inherits the objects of the one before it. That is deliberate: it lets the book look at the same data two ways, and saves printing the same function twice.
- Ch 5 → Ch 6:
Cpo - Ch 7 → Ch 8 → Ch 9:
dat2,C2iv,th,D,CL,tobs(Chapter 8 runs inside Chapter 7's session) - Ch 11 → Ch 12:
DATA,PRED,TH,OM,SG,EBE,r.fo,r.foce,cov.fo
Start a new session in between and the later chapter breaks. The boundaries are
marked in the comments of R/build.R.
The random seeds are fixed. Every snippet that simulates carries its own
set.seed(), so a rerun gives the numbers printed in the book.
En/R/snippets/chNN-*.R the same code with the comments, labels and messages in English
En/build.R runs them and makes En/output/ and En/figures/
En/output/ the English edition's frozen console output
En/figures/ the English edition's frozen figures (Latin face)
En/build.R sources R/build.R and R/_freeze.R and only redirects the three
paths and the figure font, so the two editions share one pipeline: the same
order of snippets, the same shared sessions and the same seeds. Outside the
comments and the printed strings the code is identical, token for token, and
every number in En/output/ is identical to its counterpart in output/; what
differs is the labels. The English build takes as long as the Korean one.
Why the output and the figures are committed. Normally one records the code and not what it produced. An exception is made here so that someone who has not installed R can still see the results while reading the code. The code that remakes those files is in the same repository, so nothing is taken on trust.
TDM-Vanco/ is a Shiny app that runs the TDM engine described in §13.7 of the
book. Its point is that swapping the population model's output (THETA, OMEGA,
SIGMA) is all it takes to move it to another drug.
Its only external dependency is shiny (commonmark is used to render the
markdown of the help tab, and without it the help shows as plain text).
Everything else is base R.
The screen with synthetic data (TDM-Vanco/tests/fixtures/synthetic_patient.csv)
loaded. Two observed concentrations give a MAP estimate of the individual
parameters, and the next dose is proposed with a prediction interval.
1. Without downloading anything (if you have R)
shiny::runGitHub("PKwR", "AMC-CPT", subdir = "TDM-Vanco")One line. It fetches from GitHub, unpacks into a temporary folder and starts.
2. From a clone
shiny::runApp("TDM-Vanco") # from the repository root3. In a browser, with no R at all
Open https://amc-cpt.github.io/PKwR/. No installation and no server.
GitHub cannot run R, and GitHub Pages serves static files only. But
shinylive exports the app to
WebAssembly, so the R runtime runs inside the browser and static hosting is
enough. Both shiny and commonmark, which this app uses, are in the webR
repository, so it works unchanged.
install.packages("shinylive")
shinylive::export("TDM-Vanco", "docs") # builds the static site into docs/The exported site is 66 MB, nearly all of it the R runtime. Putting that on
main would add the same weight to the default-branch tarball that
runGitHub() downloads in method 1, so it lives on a separate gh-pages
branch here (Pages source: gh-pages / (root)). The runtime is fetched on
the first visit and cached by the browser afterward.
The two differ in one way that matters: in the browser version R runs inside your own browser, so patient data you load is never transmitted anywhere. The cost is a slow first load.
If a server is wanted instead, this repository can be connected to shinyapps.io or Posit Connect Cloud.
tests/regression.R runs a regression check on the synthetic data in
tests/fixtures/ and compares it against golden_baseline.rds. No real patient
data is in this repository.
source("TDM-Vanco/tests/regression.R")『약동학 with R』 (Pharmacokinetics with R), Kyun-Seop Bae (University of Ulsan College of Medicine · Asan Medical Center). Three parts, seventeen chapters. The book is published separately.
For the solutions to the exercises, or to report an error, write to ksbae@acr.kr.
The R code in this repository and what it produces are under GPL-3 (LICENSE). The text of the book and its figure captions are not included here and are separately reserved.
| 1 Scientific Computation with R | https://github.com/AMC-CPT/SciCompR |
| 2 Scientific Inference in Clinical Trials with R | https://github.com/AMC-CPT/CTDA |
| 4 Pharmacometrics with NONMEM and R | https://github.com/AMC-CPT/PMx |
| 5 Essentials of Clinical Drug Development (online appendix) | https://github.com/AMC-CPT/CDD |