A comparative study of in vitro dose-response estimation under extreme observations

Area of Science:

• Biomedical research
• Life sciences
• Pharmacology
• Biostatistics

Background:

• Accurate dose-response relationship estimation is critical for drug development.
• Standard median-effect equation methods are sensitive to outliers in experimental data.
• Various R packages exist for dose-response curve estimation, primarily using nonlinear least squares.

Purpose of the Study:

• To compare the robustness and efficiency of 14 different R tools for estimating dose-response curves.
• To identify reliable methods for handling extreme observations in dose-response data.
• To evaluate newly introduced beta regression-based methods against established techniques.

Main Methods:

• Comparative study of 14 R dose-response estimation tools.
• Monte Carlo simulation under comprehensive scenarios.
• Evaluation of nonlinear least squares and beta regression-based methods.
• Assessment of penalized beta regression using the mgcv package.

Main Results:

• Penalized beta regression demonstrated superior performance.
• The mgcv package provided stable and accurate estimations.
• Reliable uncertainty quantification was achieved with penalized beta regression.
• Other methods showed vulnerability to extreme observations.

Conclusions:

• Penalized beta regression, particularly with the mgcv package, is the recommended method for robust dose-response curve estimation.
• This approach effectively handles nonnormality, heteroscedasticity, and asymmetry.
• It provides reliable estimation and uncertainty quantification, outperforming traditional methods in the presence of outliers.