A statistical framework for assessing pharmacological responses and biomarkers using uncertainty estimates

Area of Science:

• Computational biology
• Pharmacogenomics
• Biomarker discovery

Background:

• High-throughput drug screening uses summary statistics from dose-response curves.
• This approach often overlooks experimental uncertainty and raw data variability.
• Accurate quantification is crucial for identifying effective treatments and biomarkers.

Purpose of the Study:

• To develop a novel Bayesian framework for modeling experimental variance in drug screening data.
• To leverage uncertainty estimates for improved biomarker discovery.
• To enhance precision medicine by refining drug response predictions.

Main Methods:

• Modeling experimental variance using Gaussian Processes.
• Applying a Bayesian framework to leverage uncertainty estimates for biomarker identification.
• Utilizing in vitro screening data from 265 compounds across 1074 cancer cell lines.

Main Results:

• Identified 24 clinically established drug-response biomarkers.
• Provided evidence for six novel biomarkers by focusing on low uncertainty associations.
• Validated uncertainty estimates using a separate drug screen with replicates.

Conclusions:

• The novel Bayesian framework effectively models experimental variance and improves biomarker discovery.
• The method enhances precision medicine by providing more reliable drug-response associations.
• Applicable to dose-response data even without replicates.