Hierarchical Deep Compartment Modeling: Integrating Deep Learning, SHAP Analysis, and Bayesian Inference in Julia

In this case study, we will showcase the Hierarchical Deep Compartment Modeling (HDCM) framework, which integrates deep learning with compartmental models. This framework employs a neural network to learn the functional relationships between covariates and compartmental model parameters. To enhance interpretability, we will use SHAP (SHapley Additive exPlanations) analysis to quantify the impact of covariates on specific model parameters. Additionally, the framework incorporates Bayesian inference to establish a hierarchical model structure, estimate fixed and random effects, and quantify the uncertainty around model parameters, predictions, and SHAP analysis outputs. The entire framework is implemented in Julia, leveraging its powerful capabilities for scientific computing. Participants will learn how to implement hierarchical models, apply SHAP for covariate analysis, and interpret the results to improve model performance and reliability.