Abstract
REAP-2 is an interactive dose-response curve estimation tool for Robust and Efficient Assessment of drug Potency. It provides user-friendly dose-response curve estimation for in vitro studies and conducts statistical testing for model comparisons with a redesigned user interface. We also make a major update of the underlying estimation method with penalized beta regression, which demonstrates great reliability and accuracy in dose estimation and uncertainty quantification. In this note, we describe the method and implementation of REAP-2 with a highlight on potency estimation and drug comparison.
Introduction
In drug discovery, characterizing the relationship between compound concentration and biological response via dose-response curves is fundamental. However, in vitro experiments frequently yield "extreme observations"—data points at 0% or 100% response. Standard statistical models, such as ordinary least squares, often fail in these scenarios because they assume a normal distribution of errors, which is inappropriate for data bounded between 0 and 1. Researchers typically face a dilemma: exclude these points and lose information, or include them and risk biased results. REAP-2 addresses this by providing a robust, automated framework for handling such edge cases with high precision.
Methods
REAP-2 utilizes the median-effect equation, a generalized form of the Hill equation, as its core model. To overcome the limitations of standard regression, the tool implements Penalized Beta Regression.
The Beta distribution is naturally bounded between 0 and 1, making it ideal for "fraction affected" data and accounting for heteroscedasticity. By leveraging the mgcv R package, REAP-2 applies an L2 ridge penalty to the log-likelihood function, stabilizing the model against noise and preventing overfitting. The tool is deployed as an R Shiny web application, allowing users to configure model features and plot specifics through an intuitive interface.
Implementation and Results
The effectiveness of the penalized beta regression approach was validated through extensive Monte Carlo simulations, where it outperformed 13 other common methods in accuracy and coverage probability. REAP-2 translates these statistical advantages into a user-friendly platform.
Users can upload data, visualize dose-response curves, and generate comprehensive summaries. The tool provides reliable potency metrics (e.g., $IC_{50}$, $IC_{90}$) with tighter confidence intervals than traditional methods. Furthermore, it includes automated hypothesis testing to compare the efficacy of different agents statistically.
Conclusion
REAP-2 democratizes advanced statistical modeling for the pharmacology community. By wrapping complex penalized regression into a point-and-click interface, it enables more robust and efficient drug potency assessment. Future developments will focus on expanding model flexibility and integrating tools for identifying drug-drug interactions, such as synergy or antagonism.
Summary updated