Enhancing spectroscopy analysis with machine learning

Researchers have developed and optimised a novel sensing strategy for isomer discrimination by combining advanced spectroscopy with machine learning.

The approach is based on surface-enhanced Raman scattering (SERS) substrates (ie, plasmonic platforms). Montes-García et al. explained that they fabricated advanced SERS substrates made of quasi-spherical gold nanoparticles (Au NPs) synthesised via a seeded growth method.

About the advanced technology-driven spectroscopy technique

Their method provided “rapid, ultrasensitive, and precise discrimination” and could “differentiate closely related structural isomers”. The study investigated structural isomers, geometric isomers, and optical isomers (eg, R/S-ibuprofen).

For instance, when validating their technique with commercial ibuprofen samples, the results showed “excellent agreement with traditional circular dichroism results, highlighting the method’s robustness and precision”.

Integration of machine learning algorithms “significantly” enhanced both quantitative analysis and classification accuracy, “achieving detection limits as low as 2 × 10−8m”.

Current isomer discrimination methods

While traditional methods for differentiating types of isomers, such as mass spectrometry and nuclear magnetic resonance (NMR) spectroscopy, provide detailed molecular information during isomer discrimination, they have limitations, the authors asserted.

Therefore, “novel analytical methods that allow for the rapid, unequivocal, and ultrasensitive isomer discrimination are highly sought after”, they explained.

Current SERS techniques often encounter challenges such as inconsistent nanoparticle aggregation or “difficulty in distinguishing closely related isomers, especially optical isomers, where only probes relying on differences in hydrogen bonding with the isomers have been explored”.

Positively, the researchers concluded that their strategy addressed key challenges “in the sensitivity, uniformity, and versatility of SERS substrates”.

As part of the research, the team studied five different gold nanoparticles (Au NPs) sizes and deposition conditions. Optimal SERS performance, characterised by both efficiency and uniformity, was achieved with Au NPs “of at least 55nm and two deposition cycles”, according to the authors.

Montes-García et al. summarised that integrating machine learning algorithms with SERS “not only enhances analytical performance but also paves the way for real-time and high-throughput applications”.

This paper was published in Advanced Sensor Research.