Penalized negative binomial models for modeling an overdispersed count outcome with a high-dimensional predictor...

Area of Science:

• Genomics
• Toxicology
• Biostatistics

Background:

• Micronuclei (MN) are biomarkers for genotoxic exposure and cancer risk.
• Existing MN scoring guidelines lack statistical analysis methods, leading to varied interpretations.
• Integrating MN data with high-throughput genomic technologies can reveal molecular features of micronucleation.

Purpose of the Study:

• To address the lack of statistical methods for analyzing MN frequency data, particularly in high-dimensional settings.
• To present a novel penalized negative binomial regression model for over-dispersed count data when the number of explanatory variables (P) exceeds the number of samples (N).
• To evaluate the performance of the proposed model against penalized Poisson models using simulation studies.

Main Methods:

• Developed a penalized negative binomial regression model for count data analysis where P > N.
• Conducted simulation studies to compare the new model with penalized Poisson models for over-dispersed outcomes.
• Applied the model to analyze micronuclei frequency and gene expression data from the Norwegian Mother and Child Cohort Study.

Main Results:

• The penalized negative binomial regression model effectively handles over-dispersed count data in high-dimensional settings (P > N).
• Demonstrated superior performance compared to penalized Poisson models under over-dispersion.
• Successfully applied the method to real-world MN frequency and gene expression data.

Conclusions:

• The developed penalized negative binomial regression model provides a robust statistical framework for analyzing MN frequency data, especially with high-dimensional genomic information.
• The `countgmifs` R package facilitates the application of this method to discrete outcome datasets with high-dimensional covariates.
• This approach enhances the statistical rigor for interpreting genotoxic exposure and cancer risk biomarkers.