Bayesian machine learning optimization of microneedle design for biological fluid sampling

Abstract

The deployment of microneedles in biological fluid sampling and drug delivery is an emerging field in biotechnology, which contributes greatly to minimally-invasive methods in medicine. Prior studies on microneedles proposed designs based on the optimization of physical parameters through trial-and-error method. While these methods showed adequate results, it is possible to enhance the performance of a microneedle using a large dataset of parameters and their respective performance using advanced data analysis methods. Machine Learning (ML) offers the ability to mimic human learning behavior to expedite decision-making processes in biotechnology. In this study, the finite element analysis and ML models are combined to determine the optimal physical parameters for a microneedle design to maximize the amount of collected biological fluid. The fluid behavior in a microneedle patch is modeled using COMSOL Multiphysics (R), and the model is simulated with a set of initial physical and geometrical parameters in MATLAB (R) using LiveLink (TM). The mathematical model is used as the input to MATLAB's Bayesian Optimization function (bayesopt) and optimized for the maximum volumetric flow rate with pre-defined number of iterations. Within the parameter bounds, maximum volumetric flow rate is determined to be 21.16 mL min-1, which is 60% higher with respect to a system, where geometrical parameters are chosen randomly on average. This study introduces an online method for designing microneedles, where user can define the upper and lower bounds of the parameters to obtain an optimal design. The deployment of microneedles in biological fluid sampling and drug delivery is an emerging field in biotechnology, which contributes greatly to minimally-invasive methods in medicine.