SERS and advanced chemometrics - Utilization of Siamese neural network for picomolar identification of beta-lactam antibiotics resistance gene fragment

The enormous development and expansion of antibiotic-resistant bacterial strains impel the intensive search for new methods for fast and reliable detection of antibiotic susceptibility markers. Here, we combined DNA-targeted surface functionalization, surface-enhanced Raman spectroscopy (SERS) measurements, and subsequent spectra processing by decision system (DS) for detection of a specific oligonucleotide (ODN) sequence identical to a fragment of blaNDM-1 gene, responsible for beta-lactam antibiotic resistance. The SERS signal was measured on plasmonic gold grating, functionalized with capture ODN, ensuring the binding of corresponded ODNs. Designed DS consists of a Siamese neural network (SNN) coupled with robust statistics and Bayes decision theory. The proposed approach allows manipulation with complex multicomponent samples and predefine the desired detection level of confidence and errors, automatically determining the number of required spectra and samples. In constant to commonly used classification-type SNN, our method was applied to analyze samples with compositions previously unknown to DS. The detection of targeted ODN was performed with >= 99% level of confidence up to 3 x 10(-12) M limit on the background of 10(-10) M concentration of similar but not targeted ODNs. (C) 2021 Elsevier B.V. All rights reserved.

Automated summary

This study combines DNA-targeted surface functionalization, SERS measurements, and decision system to detect antibiotic susceptibility markers with high confidence, allowing manipulation with complex multicomponent samples and automatically determining the required number of samples, achieving high confidence detection of targeted ODN.