Quantification of Analyte Concentration in the Single Molecule Regime Using Convolutional Neural Networks

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Single-molecule (SM) detection is the pinnacle of chemical analysis, with techniques like surface-enhanced Raman scattering (SERS) showing significant promise.
• SERS allows for the identification of Raman-active molecules without recognition elements and is capable of SM detection.
• Quantifying ultralow analyte concentrations using SM detection events remains challenging, often requiring extensive calibration for each molecule.

Purpose of the Study:

• To develop a robust and facile method for concentration quantification at ultralow analyte levels using SERS.
• To demonstrate the application of convolutional neural networks (CNNs) for analyzing SERS spectral data.
• To explore the utility of transfer learning for training CNN models on new analyte molecules.

Main Methods:

• Application of a convolutional neural network (CNN) model to bundles of SERS spectra.
• Utilizing transfer learning to reduce data requirements for training CNN models on novel analytes.
• Experimental validation of the method for concentration quantification down to 10 femtomolar (fM).

Main Results:

• A CNN model applied to SERS spectra provides robust quantification down to 10 fM using SM detection events.
• Transfer learning significantly decreases the amount of data needed to train CNN models for new analyte molecules.
• The developed method offers a facile approach for analyzing large spectral datasets.

Conclusions:

• CNNs coupled with SERS enable robust and facile quantification at ultralow concentrations, including SM detection.
• Transfer learning enhances the adaptability of CNN models for diverse analyte detection.
• This approach paves the way for advanced applications in metabolomics, water quality, and fundamental research.