Machine learning enabled protein secondary structure characterization using drop-coating deposition Raman...

Area of Science:

• Biophysical Chemistry
• Spectroscopy
• Protein Science

Background:

• Protein structure characterization is crucial for therapeutic protein development.
• Accurate secondary structure analysis informs drug efficacy and stability.
• Existing methods can be time-consuming or costly.

Purpose of the Study:

• To evaluate Drop-Coating Deposition Raman (DCDR) spectroscopy for rapid protein secondary structure analysis.
• To develop and validate a machine learning model for predicting protein structural composition from DCDR spectra.
• To assess the accuracy of DCDR spectroscopy compared to established methods.

Main Methods:

• Proteins were analyzed using Drop-Coating Deposition Raman (DCDR) spectroscopy.
• Raman spectra in the amide I and II regions were analyzed using peak fitting.
• Partial Least Squares (PLS) regression modeling was employed for machine learning analysis.
• Model performance was validated on independent datasets, including IgG proteins.

Main Results:

• DCDR spectroscopy provided high-resolution, signal-rich spectra for protein secondary structure analysis.
• Peak fitting accurately quantified six secondary structure motifs (α-helices, 3₁₀-helices, β-sheets, turns, bends, random coil).
• PLS regression achieved low prediction errors for all six structural components (e.g., α-Helix: 1.36%, β-Sheet: 0.78%).
• The PLS model accurately predicted secondary structures in independent IgG protein samples (e.g., α-Helix: 3.1%, β-Sheet: 2.3%).

Conclusions:

• DCDR spectroscopy is a rapid, cost-effective technique for protein secondary structure characterization.
• Machine learning, specifically PLS regression, significantly enhances the predictive power of DCDR spectral data.
• This combined approach offers accuracy comparable to X-ray crystallography for structural motif estimation, facilitating therapeutic protein development.