Machine Learning-Driven Surrogate Modeling and Operating-Point Selection for a Microfluidic Diffusion-Membrane...

Area of Science:

• Biomedical Engineering
• Materials Science
• Computational Science

Background:

• Microfluidic diffusion systems are valuable for in vitro transdermal drug delivery (TDD) evaluation.
• Predictive accuracy is limited by small datasets and complex transport behaviors.
• Machine learning (ML) integration offers a solution to enhance predictive capabilities.

Purpose of the Study:

• To develop accurate and generalizable ML-based surrogate models for TDD.
• To predict cumulative drug permeation under varied hydrodynamic and membrane conditions.
• To create a framework for accelerated TDD system design and optimization.

Main Methods:

• Integrated ML (SVR, MLP, RFR, GBR, XGB, KNN) with microfluidic experiments.
• Examined caffeine cream permeation across 12 device-membrane configurations.
• Employed domain-bounded Gaussian-noise augmentation for data enhancement.

Main Results:

• Support Vector Regression (SVR) demonstrated superior performance (test R² > 0.97).
• Accurately modeled nonlinear time-flow-cumulative mass TDD behavior.
• Alginate membranes showed the highest cumulative mass (~448 µg/cm²).

Conclusions:

• ML-augmented microfluidic TDD framework provides accurate predictions from minimal data.
• Derived polynomial equations enable rapid design-space exploration and analytical predictions.
• This approach accelerates the development and screening of next-generation TDD platforms.