Distinguishing different proteins based on terahertz spectra by visual geometry group 16 neural network

Area of Science:

• Biophysics
• Spectroscopy
• Artificial Intelligence

Background:

• Accurate protein detection is crucial for medical diagnostics and biological research.
• Existing methods for protein identification can be time-consuming or lack precision.
• Terahertz (THz) spectroscopy offers a non-invasive method for analyzing biological molecules.

Purpose of the Study:

• To develop an intelligent system for accurate identification of four specific proteins: albumin, collagen, pepsin, and pancreatin.
• To leverage the capabilities of the Visual Geometry Group 16 (VGG-16) neural network for protein classification.
• To explore the utility of combining THz spectral data with advanced machine learning techniques for biosensing applications.

Main Methods:

• Terahertz (THz) absorption and refractive index spectra were collected for four proteins.
• Spectral data was transformed into 2D image features using the Grassia angular summation field (GASF) method.
• The VGG-16 neural network model was trained and validated using the generated image dataset.

Main Results:

• The VGG-16 model achieved a high accuracy of 98.8% in distinguishing between the four proteins.
• Comparative analysis showed superior performance of the VGG-16 model over other machine learning models.
• The GASF method effectively converted spectral data into discriminative image features for the neural network.

Conclusions:

• The VGG-16 neural network transfer learning technique provides a rapid and accurate method for protein identification.
• This approach holds significant potential for applications in biotechnology, including biosensors, biopharmaceuticals, and medical diagnostics.
• Combining THz spectroscopy with deep learning offers a promising avenue for advanced molecular analysis.