PyData Chicago July 2024 Meetup | Bayesian Regression Links Biochemical and Cellular Efficacy

A Bayesian regression model was developed to analyze biphasic curves in drug discovery, specifically for coronavirus protease inhibitors

The model successfully linked biochemical assay results with cellular efficacy data, providing better predictive value than simpler metrics previously used

Key findings showed the dimer-only PIC50 was a better predictor of cellular efficacy compared to traditional IC50 measurements

The research focused on MERS coronavirus protease as a target, with implications for developing broad-spectrum antiviral drugs

Open source Python packages (NumPyro, PyMC, Matplotlib) were used to implement the Bayesian regression analysis

The model accounted for complex protein dynamics including monomer-dimer equilibrium and multiple inhibitor binding sites

Cellular assays are more expensive and dangerous than biochemical assays, making accurate early-stage predictions valuable

The work demonstrates the value of using comprehensive mechanistic models over simplified curve-fitting approaches

Results showed improved correlation metrics (Spearman rho, Kendall tau) compared to traditional analysis methods

The methodology can potentially be applied beyond this specific case to other drug discovery applications