Machine Learning-Enhanced Predictive Modeling for Arbitrary Deterministic Lateral Displacement Design and Test.

Area of Science:

• Biomedical Engineering
• Microfluidics
• Machine Learning

Background:

• Separating biological particles from liquids is crucial for clinical diagnostics, including liquid biopsies.
• Deterministic Lateral Displacement (DLD) is a key microfluidic technique for particle sorting by size.
• Traditional DLD design and testing are complex, with finite element analysis often yielding inaccurate particle trajectory predictions due to experimental variability.

Purpose of the Study:

• To develop a machine learning-enhanced method for accurate particle trajectory prediction in DLD devices.
• To improve the determination of critical particle sizes for efficient separation.
• To accelerate the design and development cycle of DLD chips for clinical applications.

Main Methods:

• Integrated finite element simulation with a microfluidic-optimized particle simulation algorithm (MOPSA).
• Employed a Random Forest machine learning model trained on a dataset of 132 experiments from 40 DLD chips.
• Validated the predictive model against three distinct DLD chip designs.

Main Results:

• The machine learning approach significantly improved the accuracy of particle trajectory predictions compared to traditional methods.
• The model demonstrated a high correlation between predicted and experimentally observed particle trajectories.
• Accurate simulation results were achieved across various DLD chip designs without requiring additional physical testing.

Conclusions:

• The developed machine learning-enhanced approach offers a faster and more accurate way to design and evaluate DLD chips.
• This method streamlines the development process for microfluidic devices used in clinical diagnostics.
• The enhanced simulation accuracy facilitates the optimization of DLD technology for applications like liquid biopsies.