arXiv (Cornell University) · 2019 · 12 citations · 0 references
MedicinePharmacologyBiological Systems ModelingPrecision MedicineSingle-cell Systems BiologyCell Population DynamicsReinforcement LearningStochastic ControlTherapeutic ResistanceAdaptive Dosing AlgorithmsAdaptive Feedback ControlCombination TherapyDeep Reinforcement LearningDrug Combination TherapyStochastic Dynamical Systems
The challenge in controlling stochastic systems in which low-probability\nevents can set the system on catastrophic trajectories is to develop a robust\nability to respond to such events without significantly compromising the\noptimality of the baseline control policy. This paper presents CelluDose, a\nstochastic simulation-trained deep reinforcement learning adaptive feedback\ncontrol prototype for automated precision drug dosing targeting stochastic and\nheterogeneous cell proliferation. Drug resistance can emerge from random and\nvariable mutations in targeted cell populations; in the absence of an\nappropriate dosing policy, emergent resistant subpopulations can proliferate\nand lead to treatment failure. Dynamic feedback dosage control holds promise in\ncombatting this phenomenon, but the application of traditional control\napproaches to such systems is fraught with challenges due to the complexity of\ncell dynamics, uncertainty in model parameters, and the need in medical\napplications for a robust controller that can be trusted to properly handle\nunexpected outcomes. Here, training on a sample biological scenario identified\nsingle-drug and combination therapy policies that exhibit a 100% success rate\nat suppressing cell proliferation and responding to diverse system\nperturbations while establishing low-dose no-event baselines. These policies\nwere found to be highly robust to variations in a key model parameter subject\nto significant uncertainty and unpredictable dynamical changes.\n