BayesCurveFit: Enhancing Curve Fitting for Early-Stage Compound Screening Using Bayesian Inference
Authors:
Preprint
Abstract
Accurate curve fitting is central to quantifying dose–response relationships in drug discovery. However, commonly used regression-based methods are ill-suited for early prescreening assays, where measurements are sparse and experimental noise can dominate. Bayesian approaches have been developed previously, but they are rarely optimized for such data-limited, noise-prone conditions. We present BayesCurveFit , a Bayesian framework specifically designed for robust dose–response inference under data scarcity. The workflow integrates calibrated initialization, stochastic optimization, adaptive posterior sampling, and probabilistic mixture modeling within a unified Bayesian pipeline, enabling reliable parameter estimation and uncertainty quantification from few observations. By modeling residuals with a Gaussian–Laplace hybrid distribution, BayesCurveFit remains robust to outliers where ordinary least squares and conventional regression methods fail. Simulation studies and real screening benchmarks demonstrate that BayesCurveFit outperforms state-of-the-art regression-based methods in recovering true dose–response relationships under limited sampling. In addition, a Bayesian measure of significance – posterior error probability – offers interpretable probabilistic confidence for response classification. Together, these features establish a general and easy-to-use Bayesian framework for analyzing dose–response data in early screening experiments.Keywords
This publication has 48 references indexed in Scilit:
- CurveCurator: a recalibrated F-statistic to assess, classify, and explore significance of dose–response curvesNature Communications, 2023
- The IUPHAR/BPS Guide to PHARMACOLOGY in 2024Nucleic Acids Research, 2023
- The next-generation Open Targets Platform: reimagined, redesigned, rebuiltNucleic Acids Research, 2022
- Bayesian statistics and modellingNature Reviews Methods Primers, 2021
- The target landscape of clinical kinase drugsScience, 2017
- Stan: A Probabilistic Programming Language2017
- Harnessing Connectivity in a Large-Scale Small-Molecule Sensitivity DatasetCancer Discovery, 2015
- Bayesian Data Analysis2013
- Posterior Error Probabilities and False Discovery Rates: Two Sides of the Same CoinJournal of Proteome Research, 2007
- Bayes FactorsJournal of the American Statistical Association, 1995