Predicting the behaviour of biological systems is challenging due to their complex nonlinear nature. This paper applies an online machine learning-based methodology to predict the output of biological systems without a priori knowledge of their dynamical models or training data. The presented online predictor is based on a class of artificial neural networks, called Radial Basis Function (RBF) network. The learning capability and low computational complexity of the presented online predictor makes it a promising approach to be utilized in real-time prediction of biological systems. The satisfactory performance of the presented method is validated by applying it to experimental data from a light induced gene-regulatory network.
Electric fields (EFs) are widely employed to promote tissue regeneration and accelerate wound healing. Despite extensive study, the cellular responses elicited by EFs are complex and not well understood. The present work focuses on cell migration—a process essential to organismal development, immune surveillance, and repair—and seeks to achieve its precise, closed-loop regulation. Effective control is impeded by (i) the nonlinear and stochastic nature of migratory dynamics and (ii) safety constraints that restrict the admissible EF magnitude. To address these challenges, we reformulate a neural network (NN) feedback controller previously developed for single-cell membrane-potential regulation and adapt it to guide population-level cell migration. A projection operator is embedded into the NN weight-update law to prevent maladaptive learning that arises when the control signal saturates at its EF limit. Numerical simulations confirm that the modified controller maintains accurate trajectory tracking under saturation and outperforms the original NN design. Finally, we demonstrate a proof-of-concept by implementing the controller in vitro to direct the electrotactic migration of naïve macrophages in 2D culture under a unidirectional EF. For the in vitro experiments, we compare performance to the standard proportional–integral–derivative (PID) controller.
Precision medicine tailors treatment in a way that accounts for variations in patient response. Treatment strategies can be determined based on factors such as genetic mutations, age, and diet. Another way of implementing precision medicine in a dynamic fashion is through bioelectronics equipped with real-time sensing and intelligent actuation. Bioelectronic devices such as ion pumps can be utilized to deliver therapeutic drugs. To be able to perform precision medicine, medical devices need to be able to deliver drugs with high precision. For this, closed-loop control is required to be able to change the treatment strategy as new information about the response and progression of the biological system is received. To this end, a sliding mode controller is utilized given its ability to perform satisfactory control actions when there is model uncertainty. The controller is used in an experiment with the goal of delivering a pre-determined dosage of fluoxetine throughout a period of time.