Optimal Drug Dosage Control Strategy of Immune Systems Using Reinforcement Learning

Abstract

In this article, a reinforcement learning-based drug dosage control strategy is developed for immune systems with input constraints and dynamic uncertainties to sustain the number of tumor and immune cells in an acceptable level. First of all, the state of the immune system and the desired number of tumor and immune cells are constructed into an augmented state to derive an augmented immune system. By designing a discounted non-quadratic performance index function, the robust tracking control problem of immune systems with uncertainties is transformed into an optimal tracking control problem of nominal immune systems and the drug dosage can be limited within the specified range. Hereafter, a reinforcement learning algorithm and a critic-only structure are adopted to acquire the approximate optimal drug dosage control strategy. Furthermore, theoretical proof reveals that the proposed reinforcement learning-based drug dosage control strategy ensures the number of tumor and immune cells reaches the preset level under limited drug dosages and model uncertainties. Finally, simulation study verifies the availability of the developed drug dosage control strategy in different growth models of tumor cell.

Just how transformative will AI/ML be for immuno-oncology?

“…This approach, though currently used infrequently in biomedical research, has been implemented to predict drug sensitivity as well as optimizing chemotherapeutic and radiotherapy doses in retrospective and simulated clinical settings as reviewed in Eckardt et al 97 98 Similar approaches are also being adapted to assess immunotherapeutic challenges such as achieving control of the balance of a patient's immune and tumor cells with respect to treatment. 99 Use of reinforcement learning has also been successfully used in the search for T cell receptor beta chain CDR3 sequences that have enhanced affinity for peptide sequences . 100 This has implications for adoptive T cell immunotherapy.…”

“… 97 98 Similar approaches are also being adapted to assess immunotherapeutic challenges such as achieving control of the balance of a patient’s immune and tumor cells with respect to treatment. 99 Use of reinforcement learning has also been successfully used in the search for T cell receptor beta chain CDR3 sequences that have enhanced affinity for peptide sequences . 100 This has implications for adoptive T cell immunotherapy.…”

Chaos synchronization of two coupled map lattice systems using safe reinforcement learning

“…Deep RL combined with the attractiveness of chaotic attractors [20] was applied to the switch between attractors and further reduced unnecessary control. An optimal pulse interaction mechanism by a RL framework [21] optimized the synchronization of pulse-coupled oscillator networks, even with the random delay and the frequency difference. More recently, deep RL was successfully used to synchronize two same or different low-dimensional chaotic systems [22].…”