Machine learning for composition analysis of ssDNA using chemical enhancement in SERS

Area of Science:

• Biochemistry
• Spectroscopy
• Materials Science

Background:

• Surface-enhanced Raman spectroscopy (SERS) offers high sensitivity for biochemical sensing and DNA analysis.
• Metal-specific chemical enhancement effects influence SERS spectra of adsorbed molecules.
• Distinct SERS spectra arise from molecules on different metal substrates, probing unique metal-molecule interactions.

Purpose of the Study:

• To detect differences in SERS spectra of single-stranded DNA (ssDNA) on gold versus silver nanorod substrates.
• To develop and train machine learning models (linear regression and neural networks) for predicting ssDNA composition.
• To evaluate the performance of models using combined spectral data from multiple metal substrates.

Main Methods:

• Adsorption of 200-base length ssDNA onto gold and silver nanorod substrates.
• Acquisition and analysis of SERS spectra from ssDNA on different metal surfaces.
• Development and training of linear regression and neural network (NN) models, incorporating Principal Component Analysis (PCA).

Main Results:

• Distinct SERS spectra were observed for ssDNA adsorbed on gold and silver substrates, highlighting metal-specific chemical enhancement.
• Combining SERS spectra from different metals as input for PCA and NN models significantly reduced detection errors compared to single-metal analysis.
• The NN model demonstrated superior performance in handling complex noise and data variations inherent in SERS signals from fabricated metal substrates.

Conclusions:

• Leveraging metal-specific SERS spectral differences enhances the accuracy of DNA composition analysis.
• Machine learning models, particularly neural networks, effectively integrate multi-metal SERS data for robust ssDNA analysis.
• This approach offers a promising pathway for sensitive and accurate biochemical sensing and DNA analysis.