ocbe-uio / bayesynergy

An R package for Bayesian semi-parametric modelling of in-vitro drug combination experiments

bayesynergy: flexible Bayesian modelling of synergistic interaction effects in in-vitro drug combination experiments

An R package for Bayesian semi-parametric modelling of in-vitro drug combination experiments

The bayesynergy package implements a Bayesian semi-parametric model for the drug combination experiment. Through an efficient implementation in Stan, the model provides estimates of the full dose-response surface with uncertainty quantification. From the posterior dose-response function, estimates of synergy and antagonism are derived that better reflect the true uncertainty in these estimates. The model handles incomplete and messy datasets, naturally includes replicates, and contains parallel processing for large drug combination screens.

Paper

Installation

Prior to installing the package, please install RStan, and make sure you can run the models there. Stan requires the configuration of a C++ toolchain, which can be tricky on some systems.

install.packages('devtools')
library(devtools)
install_github('ocbe-uio/bayesynergy', build_vignettes = T, build_opts = c("--no-resave-data", "--no-manual"))

Note that for macOS, XQuartz must be installed for some of the plotting functions to work properly.

Citation

Leiv Rønneberg, Andrea Cremaschi, Robert Hanes, Jorrit M Enserink, Manuela Zucknick, bayesynergy: flexible Bayesian modelling of synergistic interaction effects in in vitro drug combination experiments, Briefings in Bioinformatics, Volume 22, Issue 6, November 2021, bbab251, https://doi.org/10.1093/bib/bbab251

Updates

• 29-09-22: Updated to version 2.5. Added a new 'robust' likelihood, which should be less sensitive to outliers, particularly when combined with the new penalized complexity (PC) prior for the Matérn kernel hyperparameters. Finally, changed the default prior for the global noise term from an Inverse-gamma, to a half-Cauchy.
• 26-05-21: Updated to version 2.4.1, minor notation changes. Ability to calculate the Bayes Factor comparing models with and without interaction
• 29-03-21: Updated to version 2.4, some minor tweaks to priors, default settings and plotting function. Added option to set hyperparameters of observation noise
• 13-01-21: Updated to version 2.3, added new dataset for use with synergyscreen(), fixed bug with synergyscreen() filling up hard drive, added vignette
• 01-12-20: Updated to version 2.2, added additional plotting functions for the synergyscreen() function, added a heteroscedastic setting for the noise term, + some minor improvements overall.
• 14-10-20: Updated to version 2.1, removed Gibbs sampler, should now compile easier on Windows systems
• 27-08-20: Updated to version 2.0, now running on Stan!
• 03-03-20: Added support for missing values!
• 14-04-20: Added the option to estimate lower asymptotes for monotherapies

bayesynergy's Issues

Difference between bayesynergy and synergysplines

What is the difference between this package and synergysplines? Their descriptions are almost identical. If I understood correctly the latter is explicitly following the experiments "described in Cremaschi et. al (2019)", but there is no further information on this one.

webpage

make a pkgdown webpage for the package so that users can navigate the documentation

smaller chunks of plot function

Change CI solution

TravisCI has been shut off for open-source projects back in June, which is why https://travis-ci.org/ocbe-uio/bayesynergy is stuck on a commit from 4 months ago. I suggest moving the CI workflow for bayessynergy to a different system. I've been using GitHub Actions for this, and I can take care of the migration if you want.

interactive graph (shiny)