Rapid Prediction of Multidrug-Resistant Klebsiella pneumoniae through Deep Learning Analysis of SERS Spectra

Abstract

Klebsiella pneumoniae is listed by the WHO as a priority pathogen of extreme importance that can cause serious consequences in clinical settings. Due to its increasing multidrug resistance all over the world, K. pneumoniae has the potential to cause extremely difficult-to-treat infections. Therefore, rapid and accurate identification of multidrug-resistant K. pneumoniae in clinical diagnosis is important for its prevention and infection control. However, the limitations of conventional and molecular methods significantly hindered the timely diagnosis of the pathogen. As a label-free, noninvasive, and low-cost method, surface-enhanced Raman scattering (SERS) spectroscopy has been extensively studied for its application potentials in the diagnosis of microbial pathogens. In this study, we isolated and cultured 121 K. pneumoniae strains from clinical samples with different drug resistance profiles, which included polymyxin-resistant K. pneumoniae (PRKP; n = 21), carbapenem-resistant K. pneumoniae, (CRKP; n = 50), and carbapenem-sensitive K. pneumoniae (CSKP; n = 50). For each strain, a total of 64 SERS spectra were generated for the enhancement of data reproducibility, which were then computationally analyzed via the convolutional neural network (CNN). According to the results, the deep learning model CNN plus attention mechanism could achieve a prediction accuracy as high as 99.46%, with robustness score of 5-fold cross-validation at 98.87%. Taken together, our results confirmed the accuracy and robustness of SERS spectroscopy in the prediction of drug resistance of K. pneumoniae strains with the assistance of deep learning algorithms, which successfully discriminated and predicted PRKP, CRKP, and CSKP strains. IMPORTANCE This study focuses on the simultaneous discrimination and prediction of Klebsiella pneumoniae strains with carbapenem-sensitive, carbapenem-resistant, and polymyxin-resistant phenotypes. The implementation of CNN plus an attention mechanism makes the highest prediction accuracy at 99.46%, which confirms the diagnostic potential of the combination of SERS spectroscopy with the deep learning algorithm for antibacterial susceptibility testing in clinical settings.

Keywords: Klebsiella pneumoniae; Raman spectroscopy; carbapenem; deep learning; multidrug resistance; polymyxins.

FIG 1

Metallic nanoparticle structural characterization and their signal enhancement capacities in Raman scattering effect. (A) TEM of AgNPs; (B) TEM of AuNPs; (C) UV-Vis of AgNPs and AuNPs; (D) SEM of AgNPs plus K. pneumoniae; (E) SEM of AuNPs plus K. pneumoniae; (F) Raman spectra of pure liquid culture, AgNPs, AuNPs, K. pneumoniae (K.pn) plus AgNPs, and K. pneumoniae plus AuNPs.

FIG 2

Illustration of SERS spectra in a linearly dose-dependent manner. (A to C) Linear relationship between bacterial concentration and SERS signal intensity for CSKP, CRKP, and PRKP, respectively. (D to F) SERS spectra for CSKP, CRKP, and PRKP at the concentrations ranging from 5 M to 12 M (M, McFarland standard).

FIG 3

Schematic illustration of SERS spectra across surface areas and averaged SERS spectra of CSKP, CRKP, and PRKP. (A) Heatmap plot of SERS spectra for representative CSKP, CRKP, and PRKP strains showing enhancement consistency across a surface area of 100 by 100 μm². The inset on the upper right corner shows the xy map region scanned on the silicon wafer. (B) Average SERS spectra and the corresponding characteristic peaks of CSKP (n = 3,200), CRKP (n = 3,200), and PRKP (n = 1,344) isolated from clinical samples.

FIG 4

Boxplots for normalized intensities of SERS spectra for CRKP, CSKP, and PRKP at representative Raman shifts corresponding to important biological components. A P value of <0.05 is considered statistically significant.

FIG 5

Performance evaluations and confusion matrices of the two deep learning algorithms CNN and CNN-attention. (A) Learning curve and loss curve for CNN-attention model. (B) Learning curve and loss curve for CNN model. (C) Confusion matrix for CNN-attention model. (D) Confusion matrix for CNN model.

FIG 6

Heatmap of SERS spectra before and after inputting into the deep learning model. (A) Raw SERS spectra. (B) Model-processed SERS spectra. The spectral regions that played key roles in different K. pneumoniae groups were partially different with overlap regions, which also explained why the deep learning models could partially misidentify the SERS spectra.

FIG 7

Schematic illustration of how the number of iteration (epoch) influenced the classification effects of deep learning models. As the number of convolutional layers and the number of iterations continued to increase, the data of different categories changed from the original mixed state to the gradual separation. After epoch reached to 30, the recognition of the three K. pneumoniae groups by the model gradually became stable.