A Machine Learning Protocol for Predicting Protein Infrared Spectra.

Area of Science:

• Biophysics
• Computational Chemistry
• Spectroscopy

Background:

• Infrared (IR) absorption spectra serve as crucial chemical fingerprints for biomolecules.
• Determining protein secondary structure from IR spectra is challenging due to the computational expense of theoretical interpretations in dynamic environments.

Purpose of the Study:

• To develop a rapid and cost-effective machine learning protocol for predicting amide I IR spectra of proteins.
• To establish a method for linking protein spectral properties to their biological and chemical functions.

Main Methods:

• A novel machine learning protocol was developed utilizing key structural descriptors.
• The protocol rapidly predicts amide I IR spectra, correlating them with experimental data.
• The model's transferability was tested across various protein structures and conditions.

Main Results:

• The machine learning protocol accurately predicts protein amide I IR spectra, aligning well with experimental results.
• The model demonstrated transferability, enabling differentiation of protein secondary structures.
• The approach successfully probed temperature-induced atomic structure variations and monitored protein folding dynamics.

Conclusions:

• This machine learning approach provides a cost-effective alternative to traditional methods for analyzing protein IR spectra.
• The protocol facilitates the prediction of protein secondary structures, atomic variations, and folding processes.
• The developed tool enhances the understanding of the relationship between protein spectral characteristics and their biological functions.