A Sloppy approach to QSP: XAI enabling fit-for-purpose models
Abstract
Quantitative Systems Pharmacology (QSP) promises to accelerate drug development, enable personalized medicine, and improve the predictability of clinical outcomes. Realizing its full potential depends on effectively managing the complexity of the underlying mathematical models and biological systems. Here, we present and validate a novel QSP workflow grounded in the principles of sloppy modeling, offering a practical and principled strategy for building and deploying models in a QSP pipeline. Our approach begins with a literature-derived model, constructed to be as comprehensive and unbiased as possible by drawing from the collective knowledge of prior research. At the core of the workflow is the Manifold Boundary Approximation Method (MBAM), which simplifies models while preserving their predictive capacity and mechanistic interpretability. Applying MBAM as a context-specific model reduction strategy, we link the simplified representation directly to the downstream predictions of interest. The resulting reduced models are computationally efficient and well-suited to key QSP tasks, including virtual population generation, experimental design, and target discovery. We demonstrate the utility of this workflow through case studies involving the coagulation cascade and SHIV infection. Our analysis suggests several promising next steps for improving the efficacy of bNAb therapies in HIV infected patients within the context of a general-purpose QSP modeling workflow.