BayesCurveFit: Enhancing Curve Fitting for Early-Stage Compound Screening Using Bayesian Inference

Area of Science:

• Pharmacology
• Computational Biology
• Biostatistics

Background:

• Accurate dose-response quantification is crucial in drug discovery.
• Traditional regression methods struggle with sparse data and high noise in early screening assays.
• Existing Bayesian methods are often not optimized for data-limited, noise-prone conditions.

Purpose of the Study:

• Introduce BayesCurveFit, a novel Bayesian framework for robust dose-response inference.
• Address the limitations of conventional methods in early drug discovery screening.
• Enable reliable parameter estimation and uncertainty quantification from limited experimental data.

Main Methods:

• Developed BayesCurveFit, a unified Bayesian pipeline integrating calibrated initialization, stochastic optimization, and adaptive posterior sampling.
• Employed probabilistic mixture modeling and a Gaussian-Laplace hybrid distribution for residual modeling.
• Incorporated a Bayesian measure of significance (posterior error probability) for response classification.

Main Results:

• BayesCurveFit demonstrates robust parameter estimation and uncertainty quantification even with few observations.
• The Gaussian-Laplace hybrid distribution enhances robustness to outliers compared to ordinary least squares.
• Simulation studies and real screening benchmarks show BayesCurveFit outperforms state-of-the-art regression methods in recovering dose-response relationships under limited sampling.

Conclusions:

• BayesCurveFit provides a general and user-friendly Bayesian framework for analyzing dose-response data in early screening.
• The framework offers interpretable probabilistic confidence for response classification.
• This approach enhances the reliability of drug discovery prescreening assays with limited, noisy data.