.png){kind=icon}
.png){kind=icon}
Return
Rapid identification of pathogenic bacteria using data preprocessing and machine learning-augmented label-free surface-enhanced Raman scattering
.png){kind=icon}
PRE
AI
DOI：10.1016/j.snb.2024.136963.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Despite the advantages of label-free surface-enhanced Raman scattering (SERS) for bacterial detection, the effective implementation of this approach is hindered by several intrinsic limitations, such as the uniformity and spectral similarity of bacterial SERS spectra. In this study, we developed a machine learning (ML)-integrated hydrophobic SERS platform for sensitive bacterial detection and classification. The hydrophobic Si wafer was successfully fabricated through treatment with dimethyldichlorosilane (DMDCS), resulting in a high contact angle (> 100 degrees). To achieve the highest SERS signals for bacteria, the density of gold core-silver shell nanoparticles (Au@AgNPs) was adjusted. Subsequently, the SERS activity of the Au@AgNPs under the coffee-ring effect on normal Si substrates was compared with that under the local concentration effect on hydrophobic Si substrates. Notably, the SERS signal for bacteria detected from the hydrophobic SERS platform, which enabled local enrichment of bacteria and Au@AgNPs, was approximately 33 times higher than that obtained from the normal Si substrates. Moreover, this hydrophobic SERS platform was integrated with ML algorithms to enhance the accuracy of bacterial detection. By applying three data preprocessing techniques, including baseline correction, smoothing, and normalization, we achieved 100 % accuracy in classifying SERS signals for four bacterial species. Our simple and versatile hydrophobic SERS platform, combined with ML models, exhibits immense potential for label-free detection and classification of pathogenic bacteria.
Keywords:
Bacteria classification
surface-enhanced Raman scattering
local concentration
machine learning
data preprocessing
Journal
IF:
7.7
Papers:
3.3W
Citations:
12.6W