SedNet: A physics-informed operator-learning framework for rapid sedimentation velocity analytical...

Area of Science:

• Biophysics
• Analytical Chemistry
• Computational Biology

Background:

• Sedimentation velocity analytical ultracentrifugation (SV-AUC) data analysis traditionally relies on slow numerical solvers for the Lamm equation.
• Existing methods require repeated solutions, limiting experimental throughput and data interpretation speed.

Purpose of the Study:

• To develop a faster and more efficient method for analyzing SV-AUC data.
• To leverage machine learning, specifically physics-informed neural networks, to solve the Lamm equation.

Main Methods:

• Implemented machine learning's automatic differentiation within an inverse physics-informed neural network (PINN) to solve the Lamm equation.
• Developed a physics-informed deep operator network (PI-DeepONet) combined with a multi-layer perceptron (MLP) to create an end-to-end framework named SedNet.
• Enforced the Lamm equation via a physics-informed forward operator within the SedNet framework.

Main Results:

• Demonstrated accurate solutions to the Lamm equation using automatic differentiation and PINNs.
• Successfully generated synthetic SV-AUC data and inverted it into size distributions using the PI-DeepONet and MLP combination.
• Showcased SedNet's ability to analyze real SV-AUC experimental data, confirming its generalization capabilities.

Conclusions:

• SedNet provides a significantly faster and accurate approach for SV-AUC data analysis compared to traditional numerical methods.
• The framework's ability to generate and invert data makes it a versatile tool for biophysical characterization.
• SedNet shows strong potential as a widely applicable analytical tool in scientific research.