Bayesian gene set benchmark dose estimation for "omic" responses

Area of Science:

• Environmental toxicology
• Computational biology
• Statistical modeling

Background:

• Estimating toxic reference points, such as the benchmark dose (BMD), is crucial for environmental pollution regulation.
• Current toxicity assessments often use univariate methods, analyzing one gene or tissue at a time and ignoring correlations between endpoints.
• This limitation is particularly relevant in transcriptomics, where large-scale gene expression data is common.

Purpose of the Study:

• To develop a statistically principled multivariate estimation procedure for benchmark dose (BMD) in specified gene sets.
• To extend the foundational univariate BMD approach to handle correlated data in transcriptomics.
• To provide a robust method for regulatory toxicology and hypothesis generation.

Main Methods:

• Developed a multivariate estimation procedure for benchmark dose (BMD) calculation in gene sets.
• The method statistically accounts for correlations between genes within a set.
• Implemented the procedure using R and C++ (BS-BMD).

Main Results:

• The multivariate approach was illustrated using a 5-day rat study and Hallmark gene sets.
• Results were compared to existing benchmark dose (BMD) values computed by the EPA.
• The principled multivariate method offers an advancement over previous ad-hoc approaches for transcriptomic data.

Conclusions:

• The developed multivariate benchmark dose (BMD) method effectively estimates toxicity for gene sets by incorporating correlations.
• This approach provides a statistically sound extension of univariate methods into the multivariate domain of transcriptomics.
• The method has applications in regulatory toxicology and can facilitate hypothesis generation for mechanistic pathways.