SERSNet: Surface-Enhanced Raman Spectroscopy Based Biomolecule Detection Using Deep Neural Network.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Machine Learning

Background:

• Surface-Enhanced Raman Spectroscopy (SERS) faces challenges in biomolecule detection due to signal variability.
• Machine learning (ML) offers potential solutions, but lacks standardized procedures and benchmark datasets.
• Development of reliable ML models for SERS requires robust datasets and evaluation protocols.

Purpose of the Study:

• To establish a benchmark dataset for SERS-based molecule detection using Rhodamine 6G (R6G).
• To evaluate and compare the performance of various ML models for SERS data classification.
• To identify optimal preprocessing methods and ML model combinations for SERS analysis.

Main Methods:

• Generation of a SERS spectral benchmark dataset for Rhodamine 6G (R6G).
• Evaluation of multiple machine learning models for R6G molecule classification.
• Comparative analysis of different preprocessing techniques combined with ML models.
• Development and validation of a novel ML model, SERSNet.

Main Results:

• The SERS spectral benchmark dataset provides a standardized resource for R6G detection.
• Comparative study identified optimal preprocessing and ML model combinations.
• The developed SERSNet model demonstrated robust identification of R6G molecules.
• SERSNet achieved 95.9% balanced accuracy in cross-batch testing, indicating excellent generalization.

Conclusions:

• The SERS spectral benchmark dataset facilitates reproducible ML model development for SERS.
• SERSNet represents a significant advancement in accurate and reliable SERS-based molecule detection.
• The findings pave the way for improved SERS applications in various fields.