This repository houses modular workflows for statistical analyses of correlates of risk / protection and automated reporting of analytic results. It serves as a generalized suite of tools, based on the analyses originally designed for the USG Biostatistics Response Team's analysis of COVID-19 vaccine efficacy trials (archived here). See below for brief descriptions of each of the analysis modules. This repository is designed as the second part of an analytic pipeline, with the correlates processing module serving as an upstream component.
The correlates_reporting3 repo is more flexible compared to correlates_reporting2 (design doc). Reporting2 was designed such that each time the “button” was pressed, a correlates analysis was run for one endpoint/analysis population/dataset combination, e.g., correlates for the severe endpoint starting seven days post the D29 visit in the US part of the ENSEMBLE trial. A new analysis can be specific by adding a new TRIAL section and a new COR section in config.yml with minimal coding. Reporting3 is designed to facilitate implementation of more complex objectives, e.g., to assess whether a correlate is modified by naïve/non-naïve status. This is primarily achieved by pushing the logic concerning COR from _common.R into a new top layer within an analysis module, see cor_coxph_cove_boost.R for an example.
correlates_reporting3 makes some improvements over correlates_reporting2:
- Assay metadata such as LLOQs, ULOQs, llox_label etc are stored in a csv file and used in data mapping, correlates_processing, and correlates_reportings.
The basic structure hasn’t changed from the correlates_reporting2 repo.
-
config.yml Each section in the file describes one data file.
-
_common.R performs certain tasks that are shared across modules, partly to ensure that they are handled in the same way in the whole repo. The tasks include:
- Read the data into a data frame named dat
- Read assay metadata into a dataframe named assay_metadata. Define several assay-related named lists
- Define tfinal.tpeak
- It also acts as like a package and contains functions that implement shared logic such as making bootstrap samples, trichotomizing assay markers, etc. Keeping these functions in _common.R is more nimble than spinning them off to a package and faster development cycle. The one exception is the cove.boost.collapse.strata function, which is used by both reporting3 and processing repos and is contained in the kyotil package.
All analysis code are written in R. renv is used to manage package versions.
-
Guides on forking repo for development can be found in the next section.
-
After cloning the forked repo, start R in the root directory. Enter
renv::restore()to install packages, which takes a few hours, but only has to be performed once.
- To generate tables and figures to include in reports, run the following command in the module directory. For this command to work, there needs to be a Makefile in the module that specifies the action. (See cor_coxph as an example.)
make cor_coxph- To generate pdf reports for a module, run the following command in the root directory of the repository. For this command to work, there needs to be a report.Rmd in the module directory.
bash _build_chapter.sh cor_coxph- Troubleshooting: Error in file(filename, "r", encoding = encoding) : cannot open the connection
- Check if there is a .Rprofile in the home directory. Remove it if it exists
-
Getting started: see our contribution guidelines.
-
Code style guide, with some modifications; this will largely be enforcd with
styler. -
Project organization: mostly independent subdirectories, each incorporating
herefor path resolution. -
Package version control and virtual environments using
renv.
- Correlates of Risk (CoR) Analyses
cor_tabular: Tabular descriptions of correlates of risk.cor_graphical: Graphical descriptions of correlates of risk.cor_coxph: Cox proportional hazards modeling of risk.cor_threshold: Risk modeling based on correlate thresholds.cor_nonlinear: Nonlinear modeling and evaluation.cor_surrogates: Optimal surrogates analyses.
- Correlates of Protection (CoP) Analyses
cop_prinstrat: Principal stratification analyses.cop_stochastic: Stochastic risk and vaccine efficacy evaluation.cop_mediation: Correlate-mediated vaccine efficacy and risk.
The contents of this repository are distributed under the GPL-3 license. See
file LICENSE.md
for details.