Multidimensional surface-enhanced Raman scattering biosensor integrated convolutional neural networks for accurate...

Area of Science:

• Spectroscopy
• Biosensing
• Nanomaterials

Background:

• Surface-enhanced Raman spectroscopy (SERS) offers sensitive bacterial detection but is limited by spontaneous adsorption.
• Conventional label-free SERS captures restricted chemical information from bacterial surfaces.
• Developing advanced SERS substrates is crucial for comprehensive bacterial analysis.

Purpose of the Study:

• To develop a multidimensional SERS biosensor for enhanced bacterial detection.
• To improve bacterial identification accuracy by capturing diverse physicochemical interactions.
• To explore substrate surface modifications for modulated selective adsorption.

Main Methods:

• Substrate surface modification with molecular modifiers of distinct chemical characteristics.
• Characterization of nanomaterials using UV-vis spectroscopy, SEM, DLS, and zeta potential analysis.
• Construction of a large SERS spectral database and analysis using a 1D-convolutional neural network (1D-CNN).

Main Results:

• A multidimensional SERS biosensor was successfully developed.
• A database of 119,000 SERS profiles from 17 bacterial strains across seven dimensions was created.
• A 1D-CNN model achieved 99.29% accuracy in bacterial identification using 127 dimensional combinations.

Conclusions:

• Multidimensional SERS biosensors enhance bacterial identification accuracy by leveraging rich biochemical diversity.
• Modulating selective adsorption through molecular modifiers increases physicochemical interactions.
• Dimensionality optimization is necessary to address data redundancy and overfitting in SERS analysis.