DiffDose: Differentiable Programming for Personalized Dose-Regimen Optimal Control
- 1McGill University
Abstract
Dose-regimen design requires choosing how much drug to give, when to give it, and how treatment should vary across patients. Mechanistic pharmacokinetic-pharmacodynamic (PK/PD) and quantitative systems pharmacology (QSP) models can predict treatment responses, but optimizing dosing inputs depends on model-specific sensitivity derivations or derivative-free search. Here, we introduce DiffDose, a differentiable programming framework for mechanistic open-loop dose-regimen optimization that uses automatic differentiation (AD) to handle clinically interpretable dose amounts and administration times as differentiable controls. We evaluate our method in three settings: fixed-schedule dose-amplitude optimization in OptiDose PK/PD benchmarks; dose-timing optimization in a chemotherapy-induced neutropenia model with state-dependent delay; and individualized mosunetuzumab dosing in a QSP virtual population. Across these examples, AD produced gradients consistent with references, reduced model-specific derivative work, and shortened benchmark time to solution. DiffDose thereby turns mechanistic PK/PD and QSP models from tools that evaluate prespecified regimens into gradient-based engines for individualized regimen design.
Related
- Case Report: Acute liver failure in systemic lupus erythematosus with autoimmune hepatitis–primary biliary cholangitis overlap: a belimumab-containing regimen and longitudinal follow-up
- The influence of drug dosing regimen on extinction and adaptation during experimental evolution in Candida albicans
- Erlotinib in a once-daily high-dose oral regimen could mitigate Chikungunya virus titer in mice
- Complete Donor T Cell Chimerism Predicts Lower Relapse Incidence after Standard Double Umbilical Cord Blood Reduced-Intensity Conditioning Regimen Allogeneic Transplantation in Adults
- Persistent monocyte activation and remodeling in people with HIV after 36 months of switching from a 3-drug regimen to B/F/TAF or DTG/3TC: a sub-study of the DEBATE trial
BibTeX
@misc{hajhashemi2026diffdose,
title = {{DiffDose: Differentiable Programming for Personalized Dose-Regimen Optimal Control}},
author = {S. Hajhashemi and A. Emad and M. Craig},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.07.749974},
url = {https://www.biorxiv.org/content/10.64898/2026.09.07.749974v1},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - DiffDose: Differentiable Programming for Personalized Dose-Regimen Optimal Control
AU - S. Hajhashemi
AU - A. Emad
AU - M. Craig
PY - 2026
DA - 2026/09/07
PB - bioRxiv
DO - 10.64898/2026.09.07.749974
UR - https://www.biorxiv.org/content/10.64898/2026.09.07.749974v1
AB - Dose-regimen design requires choosing how much drug to give, when to give it, and how treatment should vary across patients. Mechanistic pharmacokinetic-pharmacodynamic (PK/PD) and quantitative systems pharmacology (QSP) models can predict treatment responses, but optimizing dosing inputs depends on model-specific sensitivity derivations or derivative-free search. Here, we introduce DiffDose, a differentiable programming framework for mechanistic open-loop dose-regimen optimization that uses automatic differentiation (AD) to handle clinically interpretable dose amounts and administration times as differentiable controls. We evaluate our method in three settings: fixed-schedule dose-amplitude optimization in OptiDose PK/PD benchmarks; dose-timing optimization in a chemotherapy-induced neutropenia model with state-dependent delay; and individualized mosunetuzumab dosing in a QSP virtual population. Across these examples, AD produced gradients consistent with references, reduced model-specific derivative work, and shortened benchmark time to solution. DiffDose thereby turns mechanistic PK/PD and QSP models from tools that evaluate prespecified regimens into gradient-based engines for individualized regimen design.
ER -