Physics-informed deep learning for plasmonic sensing of nanoscale protein dynamics in solution

Area of Science:

• Biophysics
• Spectroscopy
• Artificial Intelligence

Background:

• Quantifying protein secondary structure in aqueous solutions is vital for understanding protein behavior.
• Current deep learning models struggle with data limitations in aqueous environments.

Purpose of the Study:

• To develop a method for accurate, real-time quantification of nanoscale protein secondary structure in situ.
• To overcome data scarcity hindering deep learning for protein dynamics.

Main Methods:

• Integration of a mid-infrared plasmonic sensor with a synthesized complex-frequency wave (s-CFW)-informed convolutional neural network (CNN).
• Direct probing of the amide I band in sub-10-nanometer proteins.
• Utilizing s-CFW to amplify spectral features for enhanced analysis.

Main Results:

• The physics-informed CNN achieved a mean relative error of less than 0.1 in predicting secondary structure percentages.
• The developed method demonstrated over twice the accuracy of a standard CNN.
• Enabled in situ and real-time quantification of conformational changes during protein assembly.

Conclusions:

• The combined sensor and AI approach provides a powerful tool for studying protein dynamics in physiological conditions.
• Addresses critical data limitations in developing advanced deep learning models for protein science.
• Facilitates a deeper understanding of protein interactions and conformational changes.