Skip to main content
Open Access Publications from the University of California

UC Irvine

UC Irvine Electronic Theses and Dissertations bannerUC Irvine

Machine Learning for SERS Quantitative Detection of Pyocyanin

  • Author(s): Nguyen, Cuong Quoc;
  • Advisor(s): Ragan, Regina;
  • et al.
Creative Commons 'BY-ND' version 4.0 license

Since its discovery in 1977, surface-enhanced Raman scattering (SERS) spectroscopy has been cemented as a powerful spectroscopic technique. Taking advantage of local electric field enhancement from plasmonic nanostructures, SERS provides vibrational fingerprints down to the single molecule detection limits. Yet fully capitalizing on the technique has proven challenging. The problem is rooted in (1) inherent variances in SERS enhancement factors and (2) dated spectral analysis technique. To address (1), I present a fabrication scheme that produces optically uniform SERS substrates by employing electrohydrodynamic flow to drive chemical crosslinking between colloidal gold nanospheres. The resulting substrates exhibit SERS signals with relative standard deviation of 10.4 % over 100 × 100 μm2. With pyocyanin as analyte - a secondary metabolite produced by P. aeruginosa - SERS substrates exhibit limit of quantification of 1 ng·mL-1 and robust quantification of concentrations spanning 5-orders of magnitude. To address (2), I implemented three machine learning algorithms to analyze SERS spectra. Partial least squares regression is crucial in monitoring of P. aeruginosa biollm formation in a microfluidic environment, enabling detection as early as 3 h after inoculation in complex media. Feedforward artificial neural networks trained on pyocyanin data produces prediction errors of 6.2 ± 1.1 %. Finally, 1D convolutional neural networks trained with spectra stack further reduces prediction errors to an impressive 4.9 ± 0.9 %. Overall, this thesis demonstrates SERS spectroscopy as a potential diagnostic tool while laying the foundation to fully exploit its sensing capability by integrating machine learning in the analysis pipeline.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View