Deep learning prediction of patient response time course from early data via neural-pharmacokinetic/pharmacodynamic modelling

The longitudinal analysis of patient response time course following doses therapeutics is currently performed using Pharmacokinetic/Pharmacodynamic (PK/PD) methodologies, which requires significant human experience and expertise in the modeling dynamical systems. By utilizing recent advancements deep learning, we show that governing differential equations can be learnt directly from data. In particular, propose a novel neural-PK/PD framework combines key pharmacological principles with neural ordinary equations. We applied it to an drug concentration platelet clinical dataset consisting over 600 patients. model improves upon state-of-the-art respect metrics for temporal prediction. Furthermore, by incorporating PK/PD concepts into its architecture, generalize enable simulations responses untested dosing regimens. These results demonstrate potential automated predictive analytics course.