Predicting the conformations of the silk protein through deep learning

Area of Science:

• Materials Science
• Biotechnology
• Spectroscopy

Background:

• Protein conformation is crucial for silk-based material properties.
• Current methods for analyzing protein conformation from FTIR spectra are inefficient and time-consuming.

Purpose of the Study:

• To develop an efficient, accurate, and time-efficient method for evaluating silk protein conformation using FTIR spectra.
• To leverage deep learning, specifically convolutional neural networks (CNNs), for this analysis.

Main Methods:

• Development of a set of CNN-based deep learning models.
• Training and validation of models using FTIR spectra of silk proteins.
• Comparison of CNN model performance against conventional deconvolution algorithms.

Main Results:

• CNN models demonstrated high accuracy and time-efficiency in identifying silk proteins and quantifying their conformations.
• The developed models significantly outperform conventional deconvolution algorithms.
• The approach shows potential for analyzing large FTIR datasets, including FTIR imaging.

Conclusions:

• CNN-based deep learning provides a powerful tool for rapid and accurate protein conformation analysis from FTIR spectra.
• This method facilitates efficient processing of large-scale FTIR data and offers real-time analysis capabilities.
• An open-source Python program is provided to enable researchers to utilize these CNN models for their own FTIR data analysis.