Online Program

PS36-Predicting Patient Sensitivity Using Gene-Treatment Interactions with Bayesian Shrinkage Models (301151)

*Arinjita Bhattacharyya, University of Louisville 
Riten Mitra, University of Louisville 
Subhadip Pal, University of Louisville 
Shesh N. Rai, University of Louisville 

Keywords: gene-treatment interaction,patient sensitivity, shrinkage priors,biomarkers,clinical trials

Genomic biomarkers can significantly impact in the field of therapeutic strategy and medicine through their efficient inclusion in predictive allocation models for clinical trials. They hold substantial implications, especially in the field of cancer therapeutics, where tumor heterogeneity can necessitate the introduction of subgroup-specific interventions, thus opening the door to personalized/precision medicine. To this end, a combination of individual-level demographic factors (age, sex, education, marital status) along with biomedical covariates like genomic biomarkers as covariates to predict patient sensitivity in clinical trials have been used.

A critical challenge in such model-building for cancer and genomics applications is the curse of high dimensions brought about by the high number of potential genomic signatures. To tackle this, we shall be building on a well-known Bayesian hierarchical framework with global-local shrinkage priors in logistic regression for efficient variable selection. These, in turn, will be used to allocate future subgroups of patients into appropriate treatment arms adaptively. The key feature of the approach is a representation of the likelihood using a Polya-gamma data augmentation approach that can be naturally integrated with several priors, explicitly focusing on the Horseshoe, the Dirichlet Laplace, and Double Pareto priors. Posterior inference schemes based on Gibbs sampling were developed for this allocation. Extensive simulation studies were conducted; the results show excellent predictive performance in terms of accuracies in most of the simulation scenarios. Specifically, the performance of the method was successfully validated in the context of gene-treatment interaction models that are used for assessing patient sensitivity in clinical trials with applications to some cancer genomics data. Finally, a user-friendly R package and R Shiny interface were built, ready to be disseminated among practitioners.