A hierarchical constrained density regression model for predicting cluster-level dose-response

Area of Science:

• Toxicology
• Bioinformatics
• Statistical Modeling

Background:

• Transcriptomic assays generate large datasets for chemical toxicity screening.
• Current methods analyze genes individually and lack flexibility for high exposure levels.
• Existing approaches do not share information among genes within biological pathways.

Purpose of the Study:

• To introduce Constrained Logistic Density Regression (COLDER) for simultaneous modeling of gene expression data.
• To address limitations of current statistical methods in toxicity screening.
• To develop a method that accounts for dose-response shape changes and shares pathway information.

Main Methods:

• Proposed Constrained Logistic Density Regression (COLDER) model.
• Utilized a discrete logistic stick-breaking process (LSBP) for prior assignment.
• Incorporated gene-level characteristics (e.g., pathway membership) and biologically plausible shape constraints.
• Estimated posterior distribution for benchmark dose within gene pathways.

Main Results:

• COLDER enables simultaneous modeling of expression data across multiple genes.
• The method allows information sharing among genes within the same pathways.
• Posterior distribution for benchmark dose can be directly estimated.
• Model performance evaluated through simulation and a National Toxicology Program study.

Conclusions:

• COLDER offers an improved statistical approach for analyzing high-throughput toxicity data.
• The method enhances the synthesis of large transcriptomic datasets for chemical safety.
• COLDER provides a more biologically plausible and informative analysis of dose-response relationships.