A Machine-Learning Protocol for Ultraviolet Protein-Backbone Absorption Spectroscopy under Environmental...

Area of Science:

• Biophysics
• Computational Biology
• Machine Learning

Background:

• Ultraviolet (UV) absorption spectra are crucial for protein global structure characterization.
• Theoretical interpretation of UV spectra is computationally expensive due to extensive ab initio calculations required for excited states across diverse conformations.

Purpose of the Study:

• To develop a machine learning (ML) protocol for predicting far-UV (FUV) spectra of proteins.
• To overcome the computational limitations of traditional methods for UV spectra analysis.

Main Methods:

• Implementation of a novel machine learning protocol for FUV spectra prediction.
• Comparison of ML protocol accuracy against established density functional theory (DFT) calculations.

Main Results:

• The ML protocol achieves accuracy comparable to DFT calculations for protein FUV spectra.
• The ML protocol reduces computational cost by 3-4 orders of magnitude.
• Demonstrated excellent predictive power and transferability of the ML protocol.

Conclusions:

• The developed ML protocol offers a computationally efficient and accurate alternative for FUV spectra prediction.
• This method facilitates the study of protein structural mutations and folding pathways.