novartis/xgx

Exploratory Graphics for PKPD data

xgx's Issues

Data Checking - Covariates - specify cts and cat covariates at top.

Would be better to specify the continuous and categorical data subsets at the top, directly, instead of the >=8 line.

Feature Request: add number of points in each group when plotting confidence intervals

Create Rmarkdown Tips page

There are lots of little tips I find and always forget and thought it'd be nice to compile them in a page. It could be another Resource. Pasting a bit of it below to give you an idea.

---

title: Rmarkdown Tips
author: "Andy Stein"
date: "r format(Sys.time(), '%d %B, %Y')"
output:
bookdown::html_document2:
toc: yes
toc_float: yes
code_folding: hide
bibliography: Bibliography.bib

Common Errors

• Don't name code chunks with special characters (especially underscores), because cross-referencing will not work.

Header - Default

---
title: YOUR TITLE
author: "YOUR NAME"
date: "`r format(Sys.time(), '%d %B, %Y')`"
output:
  bookdown::html_document2:
    toc: yes
    toc_float: yes
    code_folding: hide
bibliography: Bibliography.bib
---

Text

Displaying messages and warnings by default

knitr::opts_chunk$set(warning = FALSE, message = FALSE)

Number of digits to display in inline code by default

knitr::knit_hooks$set(inline = function(x) {
  x <- sprintf("%1.2", x)
  paste(x, collapse = ", ")
})

Figures

Set default size

knitr::opts_chunk$set(fig.height = 4, fig.width = 8)

How to cross-reference and add caption

Tables

data.table vs kable

Set default number of rows

Change number of rows

How to cross-reference and add caption

References

set.seed(123)

#set chunk default options
knitr::opts_chunk$set(warning = FALSE, message = FALSE, fig.height = 4, fig.width = 8)

# Output only 2 significant digits for inline r outputs
knitr::knit_hooks$set(inline = function(x) {
  x <- sprintf("%1.2g", x)
  paste(x, collapse = ", ")
})

Data Checking: update and simplify with Andy's DoseFindingToolbox IDA plots.

Meaning of steady state prof day?

Hi,

I checked the R code of "PK-Multiple Ascending Dose" and wonder what does SS_PROFDAY = 6 # steady state prof day mean? Could you give an example? Thank you for your nice work!

xgx_scale_y_percentchangelog10 issues

I found a couple issues with xgx_scale_y_percentchangelog10().

1. There aren't enough ticks between 0-100%
2. The limits option seems to have different behavior for negative and positive numbers

`
data_example = data.frame(x = 1, y = seq(-.9, 1, by = 0.1))
g = ggplot(data = data_example, aes(x = x, y = y))
g = g + geom_point()
g = g + xgx_scale_y_percentchangelog10(limits = c(-90,1))
g = g + ggtitle("aren't enough ticks between 0-100% and also, the limits setting is weird - c(-90,1)")
print(g)

g = g + xgx_scale_y_percentchangelog10(limits = c(-90,100))
g = g + ggtitle("limits aren't what I expected, and labels are weird")
print(g)
`

MMRM - Consider implementing example

Data Checking: Feedback from Alison

consider custom ggpairs page instead of corrplot
figure out how to add x-label to the missing values plot

Oncology page: update with swimmer and waterfall plots from Dose Finding Toolbox

test

Add more information to Rmarkdowns - author and date

html - indent once
next lines - indent twice

title: "CTL019B2202, CTL019B2205J, CTL019C2201 ROW PK Data Exploration"
author: "Andy Stein"
date: "r format(Sys.time(), '%d %B, %Y')"
output:
html_document:
toc: true
toc_float: true
code_folding: hide

Consider using renv and its lock file and link to that instead of running sessionInfo() at the end of each Rmarkdown

Figure out why the lines got thicker, and points got larger with latest update

Try recompiling the website with a fresh R session

exposure response

I could not reproduce the figure in the code on the website
https://opensource.nibr.com/xgx/Multiple_Ascending_Dose_PKPD_binary.html

pkpd_data_wide_plot = pkpd_data_wide %>%
mutate(PROFDAY_STR = paste("Day",PROFDAY)) %>%
filter(PROFDAY %in% c(1,3,5)) %>%
mutate(CONC_QUART = cut(CONC, quantile(CONC, na.rm=TRUE), na.rm=TRUE, include.lowest = TRUE)) %>%
group_by(CONC_QUART) %>%
mutate(CONC_MIDPOINT = median(CONC))

gg <- ggplot(data = pkpd_data_wide_plot, aes(x = CONC, y = PD))
gg <- gg + geom_jitter(aes(color = TRTACT_high2low), width = 0, height = 0.05, alpha = 0.5)
gg <- gg + geom_smooth(method = "glm", method.args = list(family=binomial(link = logit)), color = "black")
gg <- gg + xgx_stat_ci(aes(x = CONC_MIDPOINT, y = PD), distribution = "binomial", geom = "errorbar", size = 0.5)
gg <- gg + xgx_stat_ci(aes(x = CONC_MIDPOINT, y = PD), distribution = "binomial", geom = "point", shape = 0, size = 4)
gg <- gg + scale_y_continuous(labels=scales::percent)
gg <- gg + labs(x = conc_label, y = pd_response_label, color = trtact_label)
gg <- gg + xgx_scale_x_log10()
gg <- gg + facet_grid(~PROFDAY_STR)
print(gg)

After I plot, it miss the error bar and points on the curve, it looks like below:

switch from xgx_annotate_filenames to labs(caption = caption) for simplicity/transparency

At the top of my code, I have something like:

caption = paste(c("Parent Dir: dosefindingtoolbox/Rmarkdown",
                  "Rmarkdown: Efficacy_Data_Overview_Plots.Rmd",
                  "Output: Efficacy_Data_Overview_Plots.html",
                  paste0("Timestamp: ", Sys.time())),
                collapse = "\n")

And then when I’m building the graphic, I have this:

g = g + labs(caption = caption)

Or another idea is at the top of the code to type something like

gcaption = theme(plot.caption = element_text(hjust = 0.5)) +
      labs(caption = paste0(dirs$top.local,"\n",
                            file.path(dirs$scripts.local, Rname),"\n",
                            file.path(dirs$scripts.local, Rmd_In),"\n",
                            file.path(dirs$results.local, Rmd_Out)))

then, later on in the code you can use

g = g + gcaption

When downloading Rmd from website, the header is now missing. not sure why.

xgx_scale_y_percentchangelog10() + geom_jitter() has problems with -100%

Hey @margoal1, the -100% (-1) points disappear when using geom_jitter(), but not when using geom_point(). Compare the two plots below. I'm not sure if this is xgx's fault or ggplot2's fault, but just wanted to raise the issue.

data = data.frame(x = c(0, 0, 1, 1, 2, 2), y = c(1, 0, -1, -.5, -1, -1))
ggplot(data, aes(x = x, y = y)) + geom_point() + xgx_scale_y_percentchangelog10()
ggplot(data, aes(x = x, y = y)) + geom_jitter(height = 0, width = 0) + xgx_scale_y_percentchangelog10()

Data Checking Page

The code for detecting continuous covariates only relies on there being >= 8 unique values. But it should also be added to the check that the variable is numeric. In some cases, categorical variables can have >= 9 unique values, if it happens where the column is a string, it causes an error in teh continuous variable data processing further down.

Data Checking: add check for is Ctrough or Cmax before or after nearest dose.

Consider using ggsurvfit package instead of survminer for TTE plots #57

This package seems to play nicer with ggplot2 (and maybe nicer with xgxr too)

https://cloud.r-project.org/web/packages/ggsurvfit/index.html

Add Search box to menu bar of site.

A search box can be added with a single line in the _site.yml file, though I think it also requires filling out description meta-data for each page as well. This is just something we could explore at some point.

https://rstudio.github.io/distill/website.html#site-search

Replace/Update some of the covariate section with what's in the Dose Finding Toolbox

Some of the looking at correlations between covariates isn't really checking for issues, it's more part of initial data analysis. it might make more sense to break this into two docs and then have some slightly more thorough IDA based on Dose Finding Toolbox.

RECIST page - clean up code a bit for dose data, but also others. consider a more standard dataset.

Feature Request - Data checking for SDTM and ADaM and dzz and zk and pctmp datasets

Remove functions that we don't use/emphasize anymore from vignettes to speed up time.

For survival pages, use the broom/tidy option for better connection with ggplot2

https://broom.tidymodels.org/reference/tidy.survfit.html

Fix Hazard Ratio and add forest plots for time to event data

For the hazard ratio, the 1.2^beta is not correct. It should be exp(beta*(value2-value1)).

Also, would be good to add forest plot code, too.

Data Checking Page - switch from gather to pivot longer and maybe bug with only 1 categorical covariate