Introduction to bayesynergy • bayesynergy

The bayesynergy R package implements a Bayesian semi-parametric regression model for estimating the dose-response function of in-vitro drug combination experiments. The Bayesian framework offers full uncertainty quantification of the dose response function and any derived summary statistics, as well as natural handling of replicates and missing data. The Bayesian model is implemented in RStan, taking advantage of the efficient ‘No U-Turn Sampler’ as well as variational Bayes for quick approximations of the true posterior.

The package is further equipped with plotting functions for summarizing a drug response experiment, parallel processing for large drug combination screen, as well as plotting tools for summarizing and comparing these.

Basic usage example

To get started, simply load the package with

library(bayesynergy)

We have included an example dataset from a large drug combination screening experiment on diffuse large B-cell lymphoma. To access it, simply run

data("mathews_DLBCL")
y <- mathews_DLBCL[[1]][[1]]
x <- mathews_DLBCL[[1]][[2]]
head(cbind(y,x))

##      Viability ibrutinib ispinesib
## [1,] 1.2295618    0.0000         0
## [2,] 1.0376006    0.1954         0
## [3,] 1.1813851    0.7812         0
## [4,] 0.5882688    3.1250         0
## [5,] 0.4666700   12.5000         0
## [6,] 0.2869514   50.0000         0

where y contains the measured post-treatment viability, and x the corresponding drug concentrations of the two drugs.

To fit the model, simply run

fit <- bayesynergy(y,x)

and wait for the sampling process to complete.

The results can then be plotted to inspect the model fit e.g. for the monotherapies

plot(fit)

And similarly for the full dose-response surface in an interactive figure

Quantitative measures of drug response and synergy, alongside summaries of the model parameters can be extracted by running

summary(fit)

##                 mean  se_mean     sd      2.5%       50%  97.5% n_eff  Rhat
## la_1[1]       0.3349 0.001656 0.0718  1.69e-01  3.45e-01  0.457  1878 1.000
## la_2[1]       0.3884 0.003316 0.0578  2.09e-01  3.97e-01  0.457   303 1.012
## log10_ec50_1  0.4795 0.003704 0.1524  2.37e-01  4.51e-01  0.841  1694 1.000
## log10_ec50_2 -1.0445 0.049508 1.2312 -3.30e+00 -8.85e-01  0.489   618 1.005
## slope_1       2.0171 0.017844 0.9022  8.81e-01  1.80e+00  4.330  2556 1.000
## slope_2       1.4840 0.021399 1.0806  1.32e-01  1.21e+00  4.329  2550 1.000
## ell           3.1276 0.040183 1.6303  1.24e+00  2.78e+00  7.015  1646 1.004
## sigma_f       0.8420 0.016650 0.8033  1.68e-01  6.10e-01  2.736  2327 1.001
## s             0.0971 0.000262 0.0152  7.24e-02  9.55e-02  0.132  3361 1.000
## dss_1        33.4873 0.044498 2.9601  2.77e+01  3.35e+01 39.352  4425 0.999
## dss_2        59.4708 0.041415 2.8112  5.36e+01  5.95e+01 64.932  4607 1.000
## rVUS_f       82.7361 0.012960 0.8749  8.09e+01  8.28e+01 84.381  4557 1.001
## rVUS_p0      73.0607 0.033201 2.2228  6.85e+01  7.31e+01 77.315  4482 1.000
## VUS_Delta    -9.6754 0.037443 2.3948 -1.44e+01 -9.67e+00 -5.082  4091 1.000
## VUS_syn      -9.7227 0.036766 2.3500 -1.45e+01 -9.69e+00 -5.332  4086 1.000
## VUS_ant       0.0474 0.001892 0.1125  4.39e-06  8.28e-05  0.384  3536 1.000
## 
## log-Pseudo Marginal Likelihood (LPML) =  52.06799

For a more detailed look into the functionality of the package, please see the other vignettes.