Development of Surface Enhanced Raman Spectra coupled with Machine Learning Analysis for Differentiation of Closely Related Species within Enterobacter cloacae Complex

Development of Surface Enhanced Raman Spectra coupled with Machine Learning Analysis for Differentiation of Closely Related Species within Enterobacter cloacae Complex | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Development of Surface Enhanced Raman Spectra coupled with Machine Learning Analysis for Differentiation of Closely Related Species within Enterobacter cloacae Complex Menglan Zhou, Xuesong Xiong, Yanbing Li, Yingchun Xu, Bing Gu, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7939547/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 07 Feb, 2026 Read the published version in BMC Microbiology → Version 1 posted 11 You are reading this latest preprint version Abstract Background Enterobacter cloacae complex (ECC) is an important nosocomial pathogen and consists of multiple similar species. The taxonomy of ECC has been consecutively updated, adding to its identification difficulty. Methods A total of 92 ECC strains isolated from bloodstream infections during 2015–2020 were collected from a tertiary hospital in China. All the strains were identified by Vitek 2 Compact and Vitek MS and then subjected to whole genome sequencing (WGS) for average nucleotide identity (ANI) analysis. Surfaced-enhanced Raman spectroscopy (SERS) with a set of machine learning algorithms was applied in identifying species within ECC. Results Seven species were identified through ANI, including 28 E. hormaechei subsp. steigerwaltii , 17 E. hormaechei subsp. xiangfangensis , 12 E. cloacae , 11 each of E. hormaechei subsp. hoffmannii and E. bugandensis , seven E. kobei and six E. roggenkampii . The Vitek 2 compact indistinguishably identified all the strains as ECC and Vitek MS correctly identified one strain of E. kobei while achieving ambiguous results for all the other isolates. SERS combined with XGBoost model achieved 97.75% accuracy with an area under the ROC curve value of 0.9982 in the identification of ECC. Conclusion SERS coupled with machine learning algorithms holds a promising potential to acquire early prediction of ECC, outperforming the capabilities of other methods. surface enhanced Raman spectra machine learning Enterobacter cloacae complex rapid identification Figures Figure 1 Figure 2 Figure 3 1. Introduction Enterobacter spp. is one of the notorious “ESKAPE” bugs ( Enterococcus faecium , Staphylococcus aureus , Klebsiella pneumoniae , Acinetobacter baumannii , Pseudomonas aeruginosa , and Enterobacter species), among which the the E. cloacae complex (ECC) is of major importance [ 1 ]. Enterobacter spp. consists of closely related species that cannot typically be identified precisely by common phenotypic tests [ 2 ]. Moreover, the taxonomy of genus is complicated by the reassignment of some species to other genus. For example, E. aerogenes has been moved to genus Klebsiella [ 3 ], E. agglomerans to genus Pantoea [ 4 ], and E. sakazakii to genus Cronobacter [ 5 ]. ECC represents the most frequently isolated Enterobacter spp. in human respiratory, urinary tract, and bloodstream infections, especially in immunocompromised individuals [ 2 , 6 ]. Based on available epidemiology data, ECC has become the third major drug-resistant Enterobacteriaceae species involved in nosocomial infections after Escherichia coli and Klebsiella pneumoniae [ 7 ]. According to China Antimicrobial Surveillance Network (CHINET) data, ECC resistance to carbapenems increased from 4.8% in 2010 to 9.7% in 2022, and the NDM carbapenemase is mostly detected in China which can not be inhibited by current enzyme inhibitors [ 8 , 9 ]. The resistance to polymyxins, another resort to CRE, was also increasing, from 2% to 5% from 2019 to 2022, even higher than Escherichia and Klebsiella (usually < 2% for both) [ 8 , 9 ]. What’s more, ECC exhibited high and variable heteroresistance to polymyxins [ 10 , 11 ]. Given the diverse nature of species and resistance distribution among different ECC, it is of great significance to precisely identify ECC into species and subspecies levels. Phenotype-based identification methods such as commercial automated biochemical assays and matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS), have been commonly used in clinical microbiology laboratories, but often failing to differentiate the species within ECC. The application of whole-genome sequencing (WGS) enables the precise identification of ECC through average nucleotide identity (ANI) and DNA-DNA hybridization (DDH). However, both methods rely on WGS which is relatively high-cost and inconvenient for routine clinical labs. Surface enhanced Raman spectroscopy (SERS) is an emerging technique based on interactions between the light and chemical bonds [ 12 ]. In recent years, the potential application of SERS in bacterial pathogen detection has been extensively explored, especially in certain closely related species, such as the Shigella spp. and Escherichia coli , the Acinetobacter baumannii / calcoaceticus complex [ 13 , 14 ]. Being one of the clinical important pathogens, so far, the usage of SERS for the identification of ECC has never been investigated. In this study, we firstly applied the SERS technique combined with machine learning models for rapid and accurate discrimination of ECC. 2. Methods and materials 2.1 Strains collection and identification A total of 92 non-duplicate ECC strains were isolated from bloodstream infections between 2015 and 2020 at Peking Union Medical College Hospital (PUMCH), Beijing, China. Initial species identification was performed using the Vitek II Compact and Vitek MS systems (bioMérieux, France). Genomic DNA was then extracted and sequenced on an Illumina HiSeq platform (paired-end 150 bp reads, average insert size 350 bp). Low-quality reads (Q < 30) and adapter sequences were removed using fastp and FastQC. Clean reads were assembled using SPAdes (v3.15.5), and contigs shorter than 200 bp were discarded. The ANI between isolates and reference genomes from NCBI RefSeq was calculated with fastANI [ 15 ]. Species and subspecies boundaries were determined based on ANI thresholds of > 95% and > 98%, respectively, following Konstantinidis and Tiedje [ 15 ] [ 12 ]. 2.2 Synthesis of silver nanoparticle (AgNPs) Silver nanoparticles were prepared following the citrate reduction method described by Tang et al. [ 16 ] with minor modifications. In brief, 33.72 mg of silver nitrate (AgNO₃; Sinopharm, Beijing, China) was dissolved in 200 mL of ultrapure water and heated to boiling under continuous stirring using a magnetic stirrer (ZNCL-BS230, Shi-Ji-Hua-Ke, Beijing, China). Once the solution reached a full boil, the heat was turned off and 8 mL of 1% (w/v) sodium citrate was added dropwise while maintaining stirring at 650 rpm. The color gradually changed to pale yellow, indicating the formation of silver nanoparticles. The mixture was cooled naturally to room temperature, and the final volume was adjusted to 200 mL with ultrapure water. To remove excess reagents, 1 mL of the colloidal suspension was centrifuged at 7,000 rpm for 7 min (Eppendorf 5430 R, USA). The supernatant was discarded, and the pellet was resuspended in 100 µL of ultrapure water to obtain the final AgNP substrate. The suspension was stored in the dark at room temperature until use to prevent photodegradation or aggregation. 2.3 Measurement of SERS spectra Each ECC strain was subcultured on Columbia blood agar and incubated overnight at 35°C. A single colony was suspended in deionized water, and the turbidity was adjusted to a 2.0 McFarland standard (approximately 6 × 10⁸ CFU/mL) using a DensiCHEK Plus instrument (bioMérieux, France). Equal volumes (5 µL each) of bacterial suspension and AgNP solution were thoroughly mixed in a microcentrifuge tube and vortexed for 5 s to ensure uniform contact between cells and nanoparticles. A 5 µL aliquot of the mixture was dropped onto a polished silicon wafer and allowed to dry naturally at room temperature. Raman spectra were collected using a Renishaw InVia Reflex Raman spectrometer equipped with a 785 nm excitation laser. Measurements were performed in mapping mode (step = 10 µm, x = 5, y = 10), and 50 random points were automatically scanned per sample. Each strain was analyzed in triplicate on independent days to ensure reproducibility. The spectral range was set to 500–1800 cm⁻¹, where most biomolecular vibrations occur. 2.4 Average SERS spectra and characteristic peaks Raw SERS spectra were processed using LabSpec 6 software (HORIBA Scientific, Japan) to minimize background interference and highlight biological information. The data processing included smoothing, denoising, baseline correction, and intensity normalization. The parameters were optimized empirically to balance noise reduction and signal retention, as previously described by Tang et al. [ 12 ].For each species, average Raman spectra were generated by computing the mean intensity at every Raman shift, and a 20% standard error band was added to visualize spectral reproducibility. Deconvolution of overlapping peaks was conducted using a mixed Gaussian-Lorentzian (Voigt) fitting function to identify distinct molecular vibration bands. As the lack of a standardized mechanism or database for SERS band assignments diminishes the utility of spectral annotation, the emphasis in most label-free SERS research is therefore on interpreting bacterial spectra and exploring new methods and applications(DOI: 10.1016/j.bios.2017.02.032 ). The processed spectra and characteristic peaks were visualized using OriginPro 2023 (OriginLab, USA). 2.5 Machine learning analysis of SERS spectra Because SERS spectra contain complex, high-dimensional information, six supervised machine learning algorithms were applied to evaluate classification performance: Adaptive Boosting (AdaBoost), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Gradient Boosting (GBoost), and eXtreme Gradient Boosting (XGBoost). All analyses were conducted in Python 3.9 using the scikit-learn package (v0.21.3). Before model training, the dataset was randomly divided into training, validation, and testing sets in a ratio of 6:2:2. Sample labels were numerically encoded using the LabelEncoder and to_categorical functions. To optimize model performance, hyperparameters were tuned using grid search (GridSearchCV) with five-fold cross-validation (cv = 5). The parameter combination that achieved the highest average validation score was used to train the final model. For each algorithm, accuracy, precision, recall, F1-score, and five-fold cross-validation scores were calculated to assess performance stability. Receiver operating characteristic (ROC) curves and the corresponding area under the curve (AUC) were generated to visualize discriminative ability. Confusion matrices were constructed to compare the predicted and true categories, highlighting any misclassifications among ECC species. 3. Results 3.1 Strains identification Using a 95% and 98% ANI cutoff to define species or subspecies boundaries, all 92 ECC strains were strictly identified. E. hormaechei was the most common species detected (56/92, 60.9%), which can be subdivided into E. hormaechei subsp. steigerwaltii (28/92, 30.4%), E. hormaechei subsp. xiangfangensis (17/92, 18.5%) and E. hormaechei subsp. hoffmannii (11/92, 12.0%). The remaining four species were E. cloacae (12/92, 12.0%), E. bugandensis (11/92, 12.0%), E. kobei (7/92, 7.6%) and E. roggenkampii (6/92, 6.5%). On the contary, the Vitek 2 compact indistinguishably identified all the strains as ECC, with a confidence value ranging from 95%-99%. The Vitek MS correctly identified one strain of E. kobei with a confidence value of 99.9% while achieving ambiguous results for all the other isolates, the majority of which was misidentified as E. cloacae and E. asburiae , each with a 50.0% confidence value (Supplementary Table S1 ). 3.2 Average, deconvoluted and characteristic peaks of SERS spectra The full Raman spectra of bacteria contain morphological characteristics and physiological information of bacteria. In this study, the SERS spectra of ECC were collected separately. We computed the average Raman spectra and standard deviations of seven ECC species, thereby quantitatively revealing the overall trends and data variations among the SERS spectra of each bacterial type. As shown in Fig. 1 .A-G, the SERS spectrum repeatability of the seven ECC species was good, and the reproducibility of the SERS spectra varying within an acceptable range. However, due to the morphological and physiological similarities among the bacterial cells of the seven ECC species, we also employed deconvolution techniques to generate SERS component bands directly associated with molecular structures. As can be seen from Fig. 1 .H-N, the deconvoluted spectra are composed of a series of Voigt sub-bands, where each sub-band represents a spectral characteristic peak. By deconstructing different sub-bands, the differences among the seven ECC species are amplified. This method effectively extracts the important characteristic peaks from the Raman spectra and eliminates interference from spurious peaks. We also examined the characteristic peaks of different ECC species based on SERS spectra (Fig. 1 .O-U). Due to the high similarities within the complex, there were still some identical peaks but with significantly varied intensities. For example, the molecular vibration at 658 cm − 1 represents COO- deformation of guanine [ 17 ], appearing in both E. hormaechei subsp. xiangfangensis and E.cloacae , 730 cm − 1 represents adenine [ 18 ], appearing in E. hormaechei subsp. steigerwaltii , E. hormaechei subsp. xiangfangensis , E.cloacae , and E. bugandensis . As for the unique characteristic peaks, they represented different molecular components and vibrations. For instance, in E. hormaechei subsp. Steigerwaltii , the peak at 1,329 cm − 1 was associated with Amide III band [ 19 ], 1,450 cm − 1 was associated with C–H deformation [ 20 ], and 1,581 cm − 1 was assigned to the Ring breath Tyrosine [ 21 ]. A prominent peak at 1,331 cm − 1 of E. hormaechei subsp. hoffmannii was associated with DNA vibration [ 22 ]. The vibration observed at 1,454 cm − 1 in E.cloacae was attributed to CH2 stretching [ 23 ]. In E. kobei , the unique peak at 661 cm − 1 and 1,332 cm-1 was assigned to glutathione and CAOAC str or ring breathing/guanine respectively [ 24 , 25 ]. Based on the results above, E. hormaechei subsp. steigerwaltii can be uniquely identified based on characteristic peaks at 1329 cm − 1 , 1450 cm − 1 or 1581 cm − 1 ; E. hormaechei subsp. hoffmannii be identified based on characteristic peaks at 731 cm − 1 , 1331 cm − 1 or 1579 cm − 1 ; E.cloacae be identified based on characteristic peaks at 1454 cm − 1 ; E. roggenkampii be identified based on characteristic peaks at 660 cm − 1 , 1004 cm − 1 , 1049 cm − 1 , 1333 cm − 1 or 1456 cm − 1 ; E. kobei be identified based on characteristic peaks at 661 cm − 1 , 1332 cm − 1 . E. hormaechei subsp. xiangfangensis can be identified based on the characteristic peaks at 658 cm − 1 combined with 1330 cm − 1 or 1455 cm − 1 . E. bugandensis can be identified based on the combination of characteristic peaks at 659 cm − 1 plus 1589 cm − 1 . 3.3 Clustering analysis using Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA) Clustering analysis was used to determine whether the seven different ECC species are separable based on their clustering in the feature coordinate system. We used the clustering method named OPLS-DA, to analyze the spectra of ECC. The results without normalization revealed that the OPLS-DA algorithm exhibit a relative lower degree of fitting between the features and spectral samples of input matrix (R2X = 0.997, R2Y = 0.666) and predictive capability for unknown samples (Q2 = 0.165) (Fig. 2 A). In contrast, OPLS-DA with normalization (Fig. 2 B) can better preserve the association rules between different spectra due to prior learning. This enables the model to exhibit a relative higher degree of fitting between the features and spectral samples of input matrix (R2X = 0.980, R2Y = 1.000), indicating that normalization can effectively distinguish the spectra of different ECC species. Furthmore, it also demonstrates predictive capability for unknown samples (Q2 = 0.313). However, overlapping of spectral data points still exists. Therefore, we need to seek more advanced machine learning methods to build rapid identification models of ECC. 3.4 Comparison of supervised learning algorithms To enable quantitative prediction of ECC, we employed six machine learning algorithms to develop an optimal decision-making model. Before analyzing the SERS data, we present the results of each model's parameter combinations obtained through grid search using a score gradient plot (Supplementary Table S2). We found that the recognition accuracy of each model was improved with the combination and iteration of parameters. Subsequently, the obtained best parameter combination was inputted into each function, and the models’ performance were evaluated using five different evaluation metrics. As shown in Table 1 , the XGBoost algorithm exhibited superior performance with the highest accuracy (accuracy = 97.78%) and stability (5Fold = 97.45%). Notably, Gboost, RF, SVM, and DT also achieved satisfactory results, possibly due to these algorithms belonging to ensemble learning, which possess the ability to handle non-linear relationships and strong feature selection. However, the AdaBoost algorithm failed to differentiate different species of ECC (accuracy = 43.13%), because of the overlapping or intersecting SERS spectra of ECC in feature space, thereby affecting the feature selecting of AdaBoost. Table 1 Comparison of the prediction capabilities of six supervised machine learning models in the analysis of the SERS spectra of Enterobacter cloacae complexes. Algorithm Accuracy Precision Recall F1-score 5Fold XGBoost 97.78% 97.78% 97.76% 97.78% 97.45% GBoost 97.49% 97.49% 97.44% 97.49% 96.67% RF 91.80% 91.80% 91.83% 91.84% 92.72% SVM 90.55% 90.55% 90.56% 90.51% 89.21% DT 84.56% 84.56% 84.54% 84.55% 82.10% AdaBoost 43.13% 43.13% 44.47% 43.04% 45.56% ROC curve and confusion matrix are commonly used to assess the performance of classification models. We employed the One-vs-All strategy to plot the ROC (Fig. 3 A) curves for each model, evaluating their ability to discriminate false positive rate (FPR) and true positive rate (TPR) on the test set. The results demonstrate that XGBoost achieved the highest AUC value (AUC = 0.9982), while the remaining algorithms also exhibited performance consistent with the metrics provided above. For the optimal classification model, we used the confusion matrix (Fig. 3 B) to examine the model's performance on spectral data in detail. It can be seen that XGBoost successfully identified all ECC spectra. For the performance on E. hormaechei subsp. steigerwaltii , 1.05% of the spectra were incorrectly classified as E. hormaechei subsp. hoffmanniic . There were also 0.51%, 0.51%, 0.51% of E. hormaechei subsp. xiangfangensis spectra misidentifying as E. cloacae , E. hormaechei subsp. hoffmanniic and E , bugandensis , respectively. The average recognition accuracy of the XGBoost model was 97.75%, which further demonstrated the potential of this algorithm in distinguishing SERS spectrum of different ECC. 4. Discussion According to the China Antimicrobial Resistance Surveillance System (CARSS) data, Enterobacter spp. rank fifth in the isolation rate of Gram-negative bacteria, accounting for 3–5% of all bacterial isolates in China [ 26 ]. Enterobacter spp. isolated from clinical samples are usually reported as E. cloacae , and sometimes E. asburiae , E. hormaechei , or E. kobei by phenotypic methods, all belonging to ECC. Due to the similarities between species, the identification of ECC is often inaccurate and can be variable when repeating the test. However, the discrimination of ECC is clinically important due to the resistance pattern variations among different species. Researchers have shown that most of the carbapenem-resistant isolates were identified as E. hormaechei subsp.xiangfangensis ST171, a clone circulating globally [ 27 ]. Besides, colistin hetero-resistance was found in all or most of E. roggenkampii , E. kobei , E. chuandaensis and E. cloacae but rarely seen in E. hormaechei subspecies and E. ludwigii [ 10 ]. Fortunately, with the development of new technologies, the taxonomy of ECC has evolved over time, from phenotypic methods [ 28 ], such as Gram staining, biochemical assays, and MALDI-TOF MS, to molecular approaches, such as 16S rRNA gene sequencing [ 29 ], marker gene hsp60 sequencing [ 30 ] and WGS based ANI and DDH [ 31 , 32 ]. The species and subspecies within ECC have been assigned to a more detailed classification. To date, twelve species including E. cloacae , E. hormaechei , E. asburiae , E. cancerogenus , E. kobei , E. ludwigii , E. mori , E. nimipressuralis , E. roggenkampii , E. chengduensis , and E. bugandensis and E. soli are assigned to ECC [ 10 , 27 ]. E. hormaechei can be subsequently divided into five subspecies ( E. hormaechei subsp. steigerwaltii , subsp. oharae , subsp. xiangfangensis , subsp. hoffmannii , and subsp. hormaechei ), adding more complexities in the identification of ECC [ 33 ]. In clinical laboratories, the accurate identification of ECC species and subspecies still remains a challenge. Routine identification of ECC is mainly dependent on phenotypic characteristics by using commercialized systems, such as Vitek 2 compact and the MALDI-TOF MS technology [ 34 ], despite that both methods can only give an ambiguous result. This was also confirmed in our study that Vitek 2 compact failed to assign ECC species and subspecies and Vitek MS only correctly identified one strain of E. kobei . Molecular methods are more suitable for precisely identification of the ECC on species level. Hsp60 typing is the earliest developed and currently most widely used method for this purpose [ 30 ]. However, disadvantages are appearing due to the reclassification of species and subspecies, leading to unclassified or misclassified results by hsp60 . Recently, multi-plex real-time PCR and combination of single gene ( dnaJ ) real-time PCR plus MALDI-TOF MS for precise ECC identification have been reported [ 35 , 36 ]. These were effective for limited species including E. cloacae , E. asbuiae , E. hormaechei , E. kobei and E. ludwigii and unable to identify subspecies. So far, only WGS based ANI and DDH are reliable methods for accurate characterization of ECC species and subspecies. Considering the long-period and high-cost of WGS and the difficulty in data analysis, there is a significant requirement to seek easier and cheaper methods. In recent years, SERS coupled with machine learning algorithms has been emerging as a new technology for the rapid and accurate discrimination of various bacterial pathogens due to its strong Raman effects [ 37 ]. Different SERS substrates and machine learning algorithms have been tried during the exploration of using SERS technique as a quantitative analytical tool for bacterial pathogen diagnosis. Though not developed long, SERS has been successfully applied in the identification of multiple bacteria, like Escherichia coli , Salmonella typhimurium , Staphylococcus aureus , Staphylococcus epidermidis , Bacillus megaterium and so on [ 38 – 40 ]. Previous studies have demonstrated that simple average SERS spectral analysis plus machine learning algorithms is sufficient for discriminating biological samples based on significant variations in characteristic peaks [ 12 , 41 , 42 ]. In this study we attempted for the first time to apply SERS in the identification of ECC. Since ECC consists of very much closely related species, it is challenging to accurately distinguish the SERS spectra of different species. To overcome this limitation, deconvoluted SERS spectra were generated to identify subtle molecular vibrations, which has been successfully used to detect differences in very similar species such as Candida and Shigella [ 13 , 43 ]. On this basis, we extracted a list of characteristic peaks within ECC. Except for E. hormaechei subsp. xiangfangensis and E. bugandensis which needs to be identified based on characteristic peaks combination, the other five species all had at least one characteristic peak unique to themselves. However, despite being recognized as unique characteristic peaks, the molecular vibrations were quite close, posing high requirements in detection. Despite our efforts to minimize undesired effects on Raman spectroscopy measurements during the acquisition of SERS signals, the measured spectral signals still encompass extraneous contributions from the instrument or the sample itself. Hence, data cleaning becomes imperative to eliminate these detrimental effects [ 44 ]. Thus, we processed the spectral feature matrix with maximum and minimum normalization and baseline correction. Subsequently, the processed feature matrices were used for OPLS-DA clustering analyses. This algorithm used specific shapes and combined features to determine sample clusters. The R2X, R2Y, and Q2 metrics were used to appraise ECC species and assess the quality of SERS data. Neverthelss, the high dimensionality and similarity of the SERS spectral data still pose significant challenges for the OPLS-DA clustering algorithm to effectively distinguish different ECC species. Therefore, exploring more advanced methods for spectral data analysis is necessary. Ciloglu FU et al combined SERS and deep learning techniques for drug-resistant Staphylococcus aureus detection, achieving an accuracy of 97.66% and an AUC of 0.99 [ 17 ]. This breakthrough garnered significant attention and spurred numerous researchers to explore intelligent spectral analysis. In this study, we used normalized and baseline-corrected spectral data as input to construct six ensemble learning models. We employed the GridSearch algorithm to analyze the appropriate hyperparameters of various models [ 45 , 46 ]. The parameter combination yielding the highest final score for each model was selected to examine SERS data and identify different species of ECC in an independent dataset. Among the six models, XGBoost demonstrated the most accurate diagnosis with the highest efficiency in analyzing different species of ECC. This novel diagnostic method holds a promising potential to attain early prediction of ECC, surpassing the capabilities of existing methods. In summary, SERS coupled with machine learning algorithms showed the potential in the identification of highly similar ECC, enabling us to understand the aspect of clinical significance, epidemiology, and drug resistance of ECC at the species level. In comparison to currently available phenotypic and molecular methods, this method is rapid, accurate, and cost-efficient that is suitable for future use in routine diagnostic tests. Declarations Ethics approval and consent to participate This study was reviewed and approved by the Ethics Committee of Peking Union Medical College Hospital (Approval No. I-22PJ396). Only identified ECC isolates collected during routine clinical diagnostics were used, and no patient-related information was accessed. Therefore, the requirement for informed consent was waived by the Ethics Committee in accordance with the “Ethical Review Measures for Biomedical Research Involving Humans” (National Health and Family Planning Commission of China, 2016) and the “Regulations of the People’s Republic of China on the Administration of Human Genetic Resources” (State Council Order No. 717, 2019). This study complied with the Declaration of Helsinki. Clinical trial Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Fundings This work was supported by National Science Foundation for Young Scientists of China (82202541), Peking Union Medical College Hospital Talent Cultivation Program Category D (UHB12396) and Fundamental Research Funds for the Central Universities (3332022012). Author Contribution MZ, XX, and YL contributed equally to this work. MZ and XX designed and performed the experiments, collected the data, and carried out the machine learning analysis. YL contributed to data interpretation, microbiological identification, and manuscript preparation. YX supervised the microbiological methodology and provided critical revision of the manuscript. BG and JG conceived and supervised the study, provided resources and finalized the manuscript. All authors read and approved the final version of the manuscript. Acknowledgements None. Data Availability All whole-genome sequencing data from this study have been deposited in the GenBank under BioProject accession No. PRJNA1226973. References Miller WR, Arias CA. ESKAPE pathogens: antimicrobial resistance, epidemiology, clinical impact and therapeutics. Nat Rev Microbiol. 2024;22(10):598–616. 10.1038/s41579-024-01054-w . Mezzatesta ML, Gona F, Stefani S. Enterobacter cloacae complex: clinical impact and emerging antibiotic resistance. Future Microbiol. 2012;7(7):887–902. 10.2217/fmb.12.61 . Tindall BJ, Sutton G, Garrity GM. Enterobacter aerogenes Hormaeche and Edwards 1960 (Approved Lists 1980) and Klebsiella mobilis, Bascomb et al. 1971 (Approved Lists 1980) share the same nomenclatural type (ATCC 13048) on the Approved Lists and are homotypic synonyms, with consequences for the name Klebsiella mobilis Bascomb 1971 (Approved Lists 1980). Int J Syst Evol Microbiol. 2017;67(2):502-4; 10.1099/ijsem.0.001572 Gavini F, Mergaert J, Beji A, Mielcarek C, Izard D, Kersters K, et al. Transfer of Enterobacter agglomerans (Beijerinck 1888) Ewing and Fife 1972 to Pantoea gen. nov. as Pantoea agglomerans comb, nov. and Description of Pantoea dispersa sp. nov. Int J Syst Evol MicroBiol. 1989;39(3):337–45. https://doi.org/10.1099/00207713-39-3-337 . Iversen C, Mullane N, McCardell B, Tall BD, Lehner A, Fanning S, et al. Cronobacter gen. nov., a new genus to accommodate the biogroups of Enterobacter sakazakii, and proposal of Cronobacter sakazakii gen. nov., comb. nov., Cronobacter malonaticus sp. nov., Cronobacter turicensis sp. nov., Cronobacter muytjensii sp. nov., Cronobacter dublinensis sp. nov., Cronobacter genomospecies 1, and of three subspecies, Cronobacter dublinensis subsp. dublinensis subsp. nov., Cronobacter dublinensis subsp. lausannensis subsp. nov. and Cronobacter dublinensis subsp. lactaridi subsp. nov. Int J Syst Evol MicroBiol. 2008;58(6):1442–7. https://doi.org/10.1099/ijs.0.65577-0 . Wisplinghoff H, Bischoff T, Tallent SM, Seifert H, Wenzel RP, Edmond MB. Nosocomial bloodstream infections in US hospitals: analysis of 24,179 cases from a prospective nationwide surveillance study. Clin Infect Dis. 2004;39(3):309–17. 10.1086/421946 . Zhou K, Yu W, Cao X, Shen P, Lu H, Luo Q, et al. Characterization of the population structure, drug resistance mechanisms and plasmids of the community-associated Enterobacter cloacae complex in China. J Antimicrob Chemother. 2018;73(1):66–76. 10.1093/jac/dkx361 . Zong Z, Feng Y, McNally A. Carbapenem and Colistin Resistance in Enterobacter: Determinants and Clones. Trends Microbiol. 2021;29(6):473–6. 10.1016/j.tim.2020.12.009 . Giamarellou H. Epidemiology of infections caused by polymyxin-resistant pathogens. Int J Antimicrob Agents. 2016;48(6):614–21. 10.1016/j.ijantimicag.2016.09.025 . Fukuzawa S, Sato T, Aoki K, Yamamoto S, Ogasawara N, Nakajima C, et al. High prevalence of colistin heteroresistance in specific species and lineages of Enterobacter cloacae complex derived from human clinical specimens. Ann Clin Microbiol Antimicrob. 2023;22(1):60. 10.1186/s12941-023-00610-1 . Guerin F, Isnard C, Sinel C, Morand P, Dhalluin A, Cattoir V, et al. Cluster-dependent colistin hetero-resistance in Enterobacter cloacae complex. J Antimicrob Chemother. 2016;71(11):3058–61. 10.1093/jac/dkw260 . Tang JW, Li JQ, Yin XC, Xu WW, Pan YC, Liu QH, et al. Rapid Discrimination of Clinically Important Pathogens Through Machine Learning Analysis of Surface Enhanced Raman Spectra. Front Microbiol. 2022;13:843417. 10.3389/fmicb.2022.843417 . Liu W, Tang JW, Mou JY, Lyu JW, Di YW, Liao YL, et al. Rapid discrimination of Shigella spp. and Escherichia coli via label-free surface enhanced Raman spectroscopy coupled with machine learning algorithms. Front Microbiol. 2023;14:1101357. 10.3389/fmicb.2023.1101357 . Xiong XS, Yao LF, Luo YF, Yuan Q, Si YT, Chen J, et al. Differentiation of closely-related species within Acinetobacter baumannii-calcoaceticus complex via Raman spectroscopy: a comparative machine learning analysis. World J Microbiol Biotechnol. 2024;40(5):146. 10.1007/s11274-024-03948-6 . Konstantinidis KT, Tiedje JM. Genomic insights that advance the species definition for prokaryotes. Proc Natl Acad Sci U S A. 2005;102(7):2567–72. 10.1073/pnas.0409727102 . Tang JW, Li F, Liu X, Wang JT, Xiong XS, Lu XY, et al. Detection of Helicobacter pylori infection in human gastric fluid through surface-enhanced Raman spectroscopy coupled with machine learning algorithms. Lab Invest. 2023;100310. 10.1016/j.labinv.2023.100310 . Ciloglu FU, Caliskan A, Saridag AM, Kilic IH, Tokmakci M, Kahraman M, et al. Drug-resistant Staphylococcus aureus bacteria detection by combining surface-enhanced Raman spectroscopy (SERS) and deep learning techniques. Sci Rep. 2021;11(1):18444. 10.1038/s41598-021-97882-4 . Chen X, Tang M, Liu Y, Huang J, Liu Z, Tian H, et al. Surface-enhanced Raman scattering method for the identification of methicillin-resistant Staphylococcus aureus using positively charged silver nanoparticles. Mikrochim Acta. 2019;186(2):102. 10.1007/s00604-018-3150-6 . Wang X, Ren L, Diao Z, He Y, Zhang J, Liu M, et al. Robust Spontaneous Raman Flow Cytometry for Single-Cell Metabolic Phenome Profiling via pDEP-DLD-RFC. Adv Sci (Weinh). 2023;10(16):e2207497. 10.1002/advs.202207497 . Fan C, Hu Z, Mustapha A, Lin M. Rapid detection of food- and waterborne bacteria using surface-enhanced Raman spectroscopy coupled with silver nanosubstrates. Appl Microbiol Biotechnol. 2011;92(5):1053–61. 10.1007/s00253-011-3634-3 . Yaacob MF, Murata A, Nor NHM, Jesse FFA, Raja Yahya MFZ. Biochemical composition, morphology and antimicrobial susceptibility pattern of Corynebacterium pseudotuberculosis biofilm. J King Saud Univ - Sci. 2021;33(1):101225. https://doi.org/10.1016/j.jksus.2020.10.022 . Walter A, Marz A, Schumacher W, Rosch P, Popp J. Towards a fast, high specific and reliable discrimination of bacteria on strain level by means of SERS in a microfluidic device. Lab Chip. 2011;11(6):1013–21. 10.1039/c0lc00536c . Lu X, Rasco BA, Jabal JM, Aston DE, Lin M, Konkel ME. Investigating antibacterial effects of garlic (Allium sativum) concentrate and garlic-derived organosulfur compounds on Campylobacter jejuni by using Fourier transform infrared spectroscopy, Raman spectroscopy, and electron microscopy. Appl Environ Microbiol. 2011;77(15):5257–69. 10.1128/AEM.02845-10 . Du Y, Han D, Liu S, Sun X, Ning B, Han T, et al. Raman spectroscopy-based adversarial network combined with SVM for detection of foodborne pathogenic bacteria. Talanta. 2022;237:122901. 10.1016/j.talanta.2021.122901 . Torre-Gutierrez LG, Martinez-Zerega BE, Oseguera-Galindo DO, Aguilar-Lemarroy A, Jave-Suarez LF, Torres-Gonzalez LA, et al. Breast cancer chemotherapy treatment monitoring based on serum sample Raman spectroscopy. Lasers Med Sci. 2022;37(9):3649–59. 10.1007/s10103-022-03646-5 . Qin X, Ding L, Hao M, Li P, Hu F, Wang M. Antimicrobial resistance of clinical bacterial isolates in China: current status and trends. JAC Antimicrob Resist. 2024;6(2):dlae052. 10.1093/jacamr/dlae052 . Ganbold M, Seo J, Wi YM, Kwon KT, Ko KS. Species identification, antibiotic resistance, and virulence in Enterobacter cloacae complex clinical isolates from South Korea. Front Microbiol. 2023;14:1122691. 10.3389/fmicb.2023.1122691 . Georgiades K, Raoult D. Defining pathogenic bacterial species in the genomic era. Front Microbiol. 2010;1:151. 10.3389/fmicb.2010.00151 . Kim M, Oh HS, Park SC, Chun J. Towards a taxonomic coherence between average nucleotide identity and 16S rRNA gene sequence similarity for species demarcation of prokaryotes. Int J Syst Evol Microbiol. 2014;64(Pt 2):346–51. 10.1099/ijs.0.059774-0 . Hoffmann H, Roggenkamp A. Population genetics of the nomenspecies Enterobacter cloacae. Appl Environ Microbiol. 2003;69(9):5306–18. 10.1128/aem.69.9.5306-5318.2003 . Richter M, Rossello-Mora R. Shifting the genomic gold standard for the prokaryotic species definition. Proc Natl Acad Sci U S A. 2009;106(45):19126–31. 10.1073/pnas.0906412106 . Goris J, Konstantinidis KT, Klappenbach JA, Coenye T, Vandamme P, Tiedje JM. DNA-DNA hybridization values and their relationship to whole-genome sequence similarities. Int J Syst Evol Microbiol. 2007;57(Pt 1):81–91. 10.1099/ijs.0.64483-0 . Qiu X, Ye K, Ma Y, Zhao Q, Wang L, Yang J. Genome sequence-based species classification of Enterobacter cloacae complex: a study among clinical isolates. Microbiol Spectr. 2024;12(6):e0431223. 10.1128/spectrum.04312-23 . De Florio L, Riva E, Giona A, Dedej E, Fogolari M, Cella E, et al. MALDI-TOF MS Identification and Clustering Applied to Enterobacter Species in Nosocomial Setting. Front Microbiol. 2018;9:1885. 10.3389/fmicb.2018.01885 . Ji Y, Wang P, Xu T, Zhou Y, Chen R, Zhu H, et al. Development of a One-Step Multiplex PCR Assay for Differential Detection of Four species (Enterobacter cloacae, Enterobacter hormaechei, Enterobacter roggenkampii, and Enterobacter kobei) Belonging to Enterobacter cloacae Complex With Clinical Significance. Front Cell Infect Microbiol. 2021;11:677089. 10.3389/fcimb.2021.677089 . Pavlovic M, Konrad R, Iwobi AN, Sing A, Busch U, Huber I. A dual approach employing MALDI-TOF MS and real-time PCR for fast species identification within the Enterobacter cloacae complex. FEMS Microbiol Lett. 2012;328(1):46–53. 10.1111/j.1574-6968.2011.02479.x . Wang L, Liu W, Tang JW, Wang JJ, Liu QH, Wen PB, et al. Applications of Raman Spectroscopy in Bacterial Infections: Principles, Advantages, and Shortcomings. Front Microbiol. 2021;12:683580. 10.3389/fmicb.2021.683580 . Wang Y, Lee K, Irudayaraj J. Silver Nanosphere SERS Probes for Sensitive Identification of Pathogens. J Phys Chem C. 2010;114(39):16122–8. 10.1021/jp1015406 . Rodriguez L, Zhang Z, Wang D. Recent advances of Raman spectroscopy for the analysis of bacteria. Anal Sci Adv. 2023;4(3–4):81–95. 10.1002/ansa.202200066 . Zhou S, Lu C, Li Y, Xue L, Zhao C, Tian G, et al. Gold Nanobones Enhanced Ultrasensitive Surface-Enhanced Raman Scattering Aptasensor for Detecting Escherichia coli O157:H7. ACS Sens. 2020;5(2):588–96. 10.1021/acssensors.9b02600 . Wang L, Tang JW, Li F, Usman M, Wu CY, Liu QH, et al. Identification of Bacterial Pathogens at Genus and Species Levels through Combination of Raman Spectrometry and Deep-Learning Algorithms. Microbiol Spectr. 2022;10(6):e0258022. 10.1128/spectrum.02580-22 . Yang N, Wang Y, Wang X, Zhang F, Xiao Y, Yan B, et al. Label-Free Detection of DNA Supramolecular Structure Formation by Surface-Enhanced Raman Spectroscopy. J Phys Chem Lett. 2022;13(26):6208–14. 10.1021/acs.jpclett.2c01461 . Pezzotti G, Kobara M, Nakaya T, Imamura H, Fujii T, Miyamoto N, et al. Raman Metabolomics of Candida auris Clades: Profiling and Barcode Identification. Int J Mol Sci. 2022;23(19). 10.3390/ijms231911736 . Guo S, Popp J, Bocklitz T. Chemometric analysis in Raman spectroscopy from experimental design to machine learning-based modeling. Nat Protoc. 2021;16(12):5426–59. 10.1038/s41596-021-00620-3 . Ma ZW, Tang JW, Liu QH, Mou JY, Qiao R, Du Y, et al. Identification of geographic origins of Morus alba Linn. through surfaced enhanced Raman spectrometry and machine learning algorithms. J Biomol Struct Dyn. 2023;41(23):14285–98. 10.1080/07391102.2023.2180433 . Tang JW, Qiao R, Xiong XS, Tang BX, He YW, Yang YY, et al. Rapid discrimination of glycogen particles originated from different eukaryotic organisms. Int J Biol Macromol. 2022;222(Pt A):1027–36. 10.1016/j.ijbiomac.2022.09.233 . Additional Declarations No competing interests reported. Supplementary Files SupplementaryTables.docx Supplementary Table S1. Identification results of 92 Enterobacter cloacae complexes strains by Vitek 2 Compact, Vitek MS and whole genome sequencing. Supplementary Table S2. The Best Combination of hyperparameter for Different Machine Learning Algorithms. Cite Share Download PDF Status: Published Journal Publication published 07 Feb, 2026 Read the published version in BMC Microbiology → Version 1 posted Editorial decision: Revision requested 15 Dec, 2025 Reviews received at journal 08 Dec, 2025 Reviews received at journal 29 Nov, 2025 Reviewers agreed at journal 24 Nov, 2025 Reviewers agreed at journal 21 Nov, 2025 Reviewers agreed at journal 12 Nov, 2025 Reviewers invited by journal 03 Nov, 2025 Editor assigned by journal 02 Nov, 2025 Editor invited by journal 29 Oct, 2025 Submission checks completed at journal 28 Oct, 2025 First submitted to journal 28 Oct, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7939547","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":543737550,"identity":"2d31193e-31d9-477b-833a-c0488e9030ab","order_by":0,"name":"Menglan Zhou","email":"","orcid":"","institution":"Peking Union Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Menglan","middleName":"","lastName":"Zhou","suffix":""},{"id":543737551,"identity":"fa89014c-f41a-4a49-9e81-77732d6e7c04","order_by":1,"name":"Xuesong Xiong","email":"","orcid":"","institution":"The Fifth People’s Hospital of Huai’an","correspondingAuthor":false,"prefix":"","firstName":"Xuesong","middleName":"","lastName":"Xiong","suffix":""},{"id":543737552,"identity":"203949e5-fe31-468e-a405-e8eeb4d8bf7e","order_by":2,"name":"Yanbing Li","email":"","orcid":"","institution":"Peking Union Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yanbing","middleName":"","lastName":"Li","suffix":""},{"id":543737553,"identity":"288c35cb-3510-4956-8851-e6a2f710d822","order_by":3,"name":"Yingchun Xu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAt0lEQVRIiWNgGAWjYBCDBH5m9oMPSNMi2c6TbECaFoPzDGYSRCk1uJH87DFvjl2e8WGGNAmGGptoglokZ6SZG/NuSy42O8x42ILhWFpuAyEt/BI5bNK82w4kbjvMkHiDseEwYS1sMC2bmxkMJIjSArdlAzODEXFaJHuemRvO3ZacOOMwMJATiPGLwfHkZw/ebrNL7O8/fvDBhxobwloYBBLYEJwEgspBgP8AG2FFo2AUjIJRMLIBAIcePJMpPETqAAAAAElFTkSuQmCC","orcid":"","institution":"Peking Union Medical College Hospital","correspondingAuthor":true,"prefix":"","firstName":"Yingchun","middleName":"","lastName":"Xu","suffix":""},{"id":543737554,"identity":"6a8c7077-0671-47ba-aeff-3f6484eb77c7","order_by":4,"name":"Bing Gu","email":"","orcid":"","institution":"Guangdong Provincial People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Bing","middleName":"","lastName":"Gu","suffix":""},{"id":543737555,"identity":"c9d45ff6-dcb2-404e-a3e6-ba30b60ca61c","order_by":5,"name":"Jiansong Gu","email":"","orcid":"","institution":"The Fifth People’s Hospital of Huai’an","correspondingAuthor":false,"prefix":"","firstName":"Jiansong","middleName":"","lastName":"Gu","suffix":""}],"badges":[],"createdAt":"2025-10-24 16:22:29","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7939547/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7939547/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12866-026-04769-3","type":"published","date":"2026-02-07T15:58:53+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":96241627,"identity":"e442b173-7fa9-4fbc-a103-54014dd39644","added_by":"auto","created_at":"2025-11-19 07:11:08","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":116984,"visible":true,"origin":"","legend":"","description":"","filename":"ManuECCSERS20251028.docx","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/6fd1083a3aebd2e640d24f48.docx"},{"id":95852331,"identity":"089e2172-fb82-491d-9ba2-dfc92c4ef143","added_by":"auto","created_at":"2025-11-13 16:05:13","extension":"tif","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3408218,"visible":true,"origin":"","legend":"","description":"","filename":"Figrue1.tif","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/c5dd5bdea1a766ff2176fbd9.tif"},{"id":95852342,"identity":"20550253-6972-4d10-87bb-65d2f0df31c8","added_by":"auto","created_at":"2025-11-13 16:05:13","extension":"tif","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7392372,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.tif","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/4dfff7a01b7b00eff1053a34.tif"},{"id":95852344,"identity":"4b851f0a-5253-4889-a72d-70966ebc185e","added_by":"auto","created_at":"2025-11-13 16:05:14","extension":"tif","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4843360,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.tif","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/468c4b5d661de0db51a50d66.tif"},{"id":95852337,"identity":"448eb5e0-5614-46d2-8ac3-eb8380c2b293","added_by":"auto","created_at":"2025-11-13 16:05:13","extension":"json","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7497,"visible":true,"origin":"","legend":"","description":"","filename":"6ede46b7b8ad4ba3ba2944f7f5599606.json","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/63acd99ad476e34917024135.json"},{"id":96240287,"identity":"f43a4a35-5a7f-45f3-b2e9-db18518b1e89","added_by":"auto","created_at":"2025-11-19 07:08:44","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":34879,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTables.docx","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/467b22c884c598981d1f2743.docx"},{"id":95852340,"identity":"8e2ec831-7040-496d-8770-ea300c6ec275","added_by":"auto","created_at":"2025-11-13 16:05:13","extension":"xml","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":126954,"visible":true,"origin":"","legend":"","description":"","filename":"6ede46b7b8ad4ba3ba2944f7f55996061enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/185f15425032c56758c5579c.xml"},{"id":96241212,"identity":"a66de087-c5ed-47df-9147-24c99e515d50","added_by":"auto","created_at":"2025-11-19 07:10:23","extension":"tif","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3408218,"visible":true,"origin":"","legend":"","description":"","filename":"Figrue1.tif","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/49149abd5d1cc8aeaa8a0748.tif"},{"id":96240289,"identity":"5a7a8b3c-f48f-4205-9c0f-6fa1cbfb276e","added_by":"auto","created_at":"2025-11-19 07:08:44","extension":"tif","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7392372,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.tif","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/4c7654188f14fd366c6f0666.tif"},{"id":96240279,"identity":"76f6b99b-eef5-430f-9d48-29f5cf7000db","added_by":"auto","created_at":"2025-11-19 07:08:43","extension":"tif","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4843360,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.tif","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/83bb6d132ba8d68d15e0a4d6.tif"},{"id":95852338,"identity":"814ee5da-bd9a-458e-8304-2f47eaa797f2","added_by":"auto","created_at":"2025-11-13 16:05:13","extension":"xml","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":125353,"visible":true,"origin":"","legend":"","description":"","filename":"6ede46b7b8ad4ba3ba2944f7f55996061structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/b50101f97ee17a57fdfeaeb4.xml"},{"id":96240462,"identity":"19246c3d-cf76-48e5-bb3a-1ef64ed9da1b","added_by":"auto","created_at":"2025-11-19 07:08:56","extension":"html","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":136293,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/d6b04d0d6a8a1180756f44e2.html"},{"id":95852330,"identity":"6cdd51c6-4f1f-4dbf-b48c-0d089b30ad5f","added_by":"auto","created_at":"2025-11-13 16:05:13","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":449135,"visible":true,"origin":"","legend":"\u003cp\u003eAverage, deconvoluted and characteristic peaks of SERS spectra of \u003cem\u003eE. hormaechei \u003c/em\u003esubsp\u003cem\u003e. steigerwaltii\u003c/em\u003e,\u003cem\u003e E. hormaechei \u003c/em\u003esubsp.\u003cem\u003e xiangfangensis\u003c/em\u003e,\u003cem\u003e E. hormaechei subsp. hoffmannii\u003c/em\u003e, \u003cem\u003eE.cloacae\u003c/em\u003e,\u003cem\u003e E. bugandensis \u003c/em\u003e,\u003cem\u003e E. roggenkampii \u003c/em\u003eand \u003cem\u003eE. kobei \u003c/em\u003e. (A-G) Average SERS spectra of ECC species. (H-N) Deconvoluted SERS spectra of ECC. X-axis is Raman shift in the range of 500-1800 cm\u003csup\u003e−1\u003c/sup\u003e; Y-axis is the relative Raman intensity. a.u.: artificial unit; the grey dashed area stands for the 20% standard error band for each average SERS spectrum. (O-U) Characteristic peaks of ECC species.\u003c/p\u003e","description":"","filename":"OnlineFigrue1.png","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/193de5717e8a7215c0eb2624.png"},{"id":96241442,"identity":"ed756912-c90b-4534-92ae-3544dbfd5db6","added_by":"auto","created_at":"2025-11-19 07:10:44","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1065023,"visible":true,"origin":"","legend":"\u003cp\u003eClustering analysis of SERS spectra of\u003cstrong\u003e \u003c/strong\u003eseven ECC species. (A) Scatterplot of SERS spectra via OPLS-DA without normalization. (B) Scatterplot of SERS spectra via OPLS-DA with normalization.\u003c/p\u003e","description":"","filename":"OnlineFigure2.png","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/60deaf6b364f69240db4fac7.png"},{"id":95852334,"identity":"bbac2088-3507-4ff7-af7b-59688ddd78eb","added_by":"auto","created_at":"2025-11-13 16:05:13","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":553567,"visible":true,"origin":"","legend":"\u003cp\u003eROC curve and confusion matrix for the machine learning support vector machine model when applied to SERS spectra of ECC. (A) ROC curve. (B) Confusion matrix. The numbers in the confusion matrix represent the possibility of correctly classified (diagonal) or mis-identified (off-diagonal) spectra, respectively.\u003c/p\u003e","description":"","filename":"OnlineFigure3.png","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/72f7db1be42348b57753378c.png"},{"id":102235839,"identity":"14c9ce8c-5c27-4dfd-9887-fff99aab7006","added_by":"auto","created_at":"2026-02-09 16:17:54","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3768507,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/bcf43094-e297-4050-8c98-1ae98567ee17.pdf"},{"id":95852333,"identity":"1b893f09-45c4-4f13-a143-c4a19a0ef7d5","added_by":"auto","created_at":"2025-11-13 16:05:13","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":34879,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Table S1. Identification results of 92\u003cem\u003e Enterobacter cloacae \u003c/em\u003ecomplexes strains by Vitek 2 Compact, Vitek MS and whole genome sequencing.\u003c/p\u003e\n\u003cp\u003eSupplementary Table S2. The Best Combination of hyperparameter for Different Machine Learning Algorithms.\u003c/p\u003e","description":"","filename":"SupplementaryTables.docx","url":"https://assets-eu.researchsquare.com/files/rs-7939547/v1/62cbc5689c0bedf8c28ec4f1.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development of Surface Enhanced Raman Spectra coupled with Machine Learning Analysis for Differentiation of Closely Related Species within Enterobacter cloacae Complex","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003e\u003cem\u003eEnterobacter\u003c/em\u003e spp. is one of the notorious \u0026ldquo;ESKAPE\u0026rdquo; bugs (\u003cem\u003eEnterococcus faecium\u003c/em\u003e, \u003cem\u003eStaphylococcus aureus\u003c/em\u003e, \u003cem\u003eKlebsiella pneumoniae\u003c/em\u003e, \u003cem\u003eAcinetobacter baumannii\u003c/em\u003e, \u003cem\u003ePseudomonas aeruginosa\u003c/em\u003e, and \u003cem\u003eEnterobacter\u003c/em\u003e species), among which the the \u003cem\u003eE. cloacae\u003c/em\u003e complex (ECC) is of major importance [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. \u003cem\u003eEnterobacter\u003c/em\u003e spp. consists of closely related species that cannot typically be identified precisely by common phenotypic tests [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Moreover, the taxonomy of genus is complicated by the reassignment of some species to other genus. For example, \u003cem\u003eE. aerogenes\u003c/em\u003e has been moved to genus \u003cem\u003eKlebsiella\u003c/em\u003e [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], \u003cem\u003eE. agglomerans\u003c/em\u003e to genus \u003cem\u003ePantoea\u003c/em\u003e [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e], and \u003cem\u003eE. sakazakii\u003c/em\u003e to genus \u003cem\u003eCronobacter\u003c/em\u003e [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. ECC represents the most frequently isolated \u003cem\u003eEnterobacter\u003c/em\u003e spp. in human respiratory, urinary tract, and bloodstream infections, especially in immunocompromised individuals [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Based on available epidemiology data, ECC has become the third major drug-resistant Enterobacteriaceae species involved in nosocomial infections after \u003cem\u003eEscherichia coli\u003c/em\u003e and \u003cem\u003eKlebsiella pneumoniae\u003c/em\u003e [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. According to China Antimicrobial Surveillance Network (CHINET) data, ECC resistance to carbapenems increased from 4.8% in 2010 to 9.7% in 2022, and the NDM carbapenemase is mostly detected in China which can not be inhibited by current enzyme inhibitors [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. The resistance to polymyxins, another resort to CRE, was also increasing, from 2% to 5% from 2019 to 2022, even higher than \u003cem\u003eEscherichia\u003c/em\u003e and \u003cem\u003eKlebsiella\u003c/em\u003e (usually\u0026thinsp;\u0026lt;\u0026thinsp;2% for both) [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. What\u0026rsquo;s more, ECC exhibited high and variable heteroresistance to polymyxins [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eGiven the diverse nature of species and resistance distribution among different ECC, it is of great significance to precisely identify ECC into species and subspecies levels. Phenotype-based identification methods such as commercial automated biochemical assays and matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS), have been commonly used in clinical microbiology laboratories, but often failing to differentiate the species within ECC. The application of whole-genome sequencing (WGS) enables the precise identification of ECC through average nucleotide identity (ANI) and DNA-DNA hybridization (DDH). However, both methods rely on WGS which is relatively high-cost and inconvenient for routine clinical labs. Surface enhanced Raman spectroscopy (SERS) is an emerging technique based on interactions between the light and chemical bonds [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. In recent years, the potential application of SERS in bacterial pathogen detection has been extensively explored, especially in certain closely related species, such as the \u003cem\u003eShigella\u003c/em\u003e spp. and \u003cem\u003eEscherichia coli\u003c/em\u003e, the \u003cem\u003eAcinetobacter baumannii\u003c/em\u003e/\u003cem\u003ecalcoaceticus\u003c/em\u003e complex [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eBeing one of the clinical important pathogens, so far, the usage of SERS for the identification of ECC has never been investigated. In this study, we firstly applied the SERS technique combined with machine learning models for rapid and accurate discrimination of ECC.\u003c/p\u003e"},{"header":"2. Methods and materials","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Strains collection and identification\u003c/h2\u003e\u003cp\u003eA total of 92 non-duplicate ECC strains were isolated from bloodstream infections between 2015 and 2020 at Peking Union Medical College Hospital (PUMCH), Beijing, China. Initial species identification was performed using the Vitek II Compact and Vitek MS systems (bioM\u0026eacute;rieux, France). Genomic DNA was then extracted and sequenced on an Illumina HiSeq platform (paired-end 150 bp reads, average insert size 350 bp). Low-quality reads (Q\u0026thinsp;\u0026lt;\u0026thinsp;30) and adapter sequences were removed using fastp and FastQC. Clean reads were assembled using SPAdes (v3.15.5), and contigs shorter than 200 bp were discarded. The ANI between isolates and reference genomes from NCBI RefSeq was calculated with fastANI [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Species and subspecies boundaries were determined based on ANI thresholds of \u0026gt;\u0026thinsp;95% and \u0026gt;\u0026thinsp;98%, respectively, following Konstantinidis and Tiedje [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Synthesis of silver nanoparticle (AgNPs)\u003c/h2\u003e\u003cp\u003eSilver nanoparticles were prepared following the citrate reduction method described by Tang \u003cem\u003eet al.\u003c/em\u003e [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] with minor modifications. In brief, 33.72 mg of silver nitrate (AgNO₃; Sinopharm, Beijing, China) was dissolved in 200 mL of ultrapure water and heated to boiling under continuous stirring using a magnetic stirrer (ZNCL-BS230, Shi-Ji-Hua-Ke, Beijing, China). Once the solution reached a full boil, the heat was turned off and 8 mL of 1% (w/v) sodium citrate was added dropwise while maintaining stirring at 650 rpm. The color gradually changed to pale yellow, indicating the formation of silver nanoparticles. The mixture was cooled naturally to room temperature, and the final volume was adjusted to 200 mL with ultrapure water.\u003c/p\u003e\u003cp\u003eTo remove excess reagents, 1 mL of the colloidal suspension was centrifuged at 7,000 rpm for 7 min (Eppendorf 5430 R, USA). The supernatant was discarded, and the pellet was resuspended in 100 \u0026micro;L of ultrapure water to obtain the final AgNP substrate. The suspension was stored in the dark at room temperature until use to prevent photodegradation or aggregation.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Measurement of SERS spectra\u003c/h2\u003e\u003cp\u003eEach ECC strain was subcultured on Columbia blood agar and incubated overnight at 35\u0026deg;C. A single colony was suspended in deionized water, and the turbidity was adjusted to a 2.0 McFarland standard (approximately 6 \u0026times; 10⁸ CFU/mL) using a DensiCHEK Plus instrument (bioM\u0026eacute;rieux, France). Equal volumes (5 \u0026micro;L each) of bacterial suspension and AgNP solution were thoroughly mixed in a microcentrifuge tube and vortexed for 5 s to ensure uniform contact between cells and nanoparticles.\u003c/p\u003e\u003cp\u003eA 5 \u0026micro;L aliquot of the mixture was dropped onto a polished silicon wafer and allowed to dry naturally at room temperature. Raman spectra were collected using a Renishaw InVia Reflex Raman spectrometer equipped with a 785 nm excitation laser. Measurements were performed in mapping mode (step\u0026thinsp;=\u0026thinsp;10 \u0026micro;m, x\u0026thinsp;=\u0026thinsp;5, y\u0026thinsp;=\u0026thinsp;10), and 50 random points were automatically scanned per sample. Each strain was analyzed in triplicate on independent days to ensure reproducibility. The spectral range was set to 500\u0026ndash;1800 cm⁻\u0026sup1;, where most biomolecular vibrations occur.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.4 Average SERS spectra and characteristic peaks\u003c/h2\u003e\u003cp\u003eRaw SERS spectra were processed using LabSpec 6 software (HORIBA Scientific, Japan) to minimize background interference and highlight biological information. The data processing included smoothing, denoising, baseline correction, and intensity normalization. The parameters were optimized empirically to balance noise reduction and signal retention, as previously described by Tang \u003cem\u003eet al.\u003c/em\u003e [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].For each species, average Raman spectra were generated by computing the mean intensity at every Raman shift, and a 20% standard error band was added to visualize spectral reproducibility. Deconvolution of overlapping peaks was conducted using a mixed Gaussian-Lorentzian (Voigt) fitting function to identify distinct molecular vibration bands. As the lack of a standardized mechanism or database for SERS band assignments diminishes the utility of spectral annotation, the emphasis in most label-free SERS research is therefore on interpreting bacterial spectra and exploring new methods and applications(DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.bios.2017.02.032\u003c/span\u003e\u003cspan address=\"10.1016/j.bios.2017.02.032\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). The processed spectra and characteristic peaks were visualized using OriginPro 2023 (OriginLab, USA).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.5 Machine learning analysis of SERS spectra\u003c/h2\u003e\u003cp\u003eBecause SERS spectra contain complex, high-dimensional information, six supervised machine learning algorithms were applied to evaluate classification performance: Adaptive Boosting (AdaBoost), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Gradient Boosting (GBoost), and eXtreme Gradient Boosting (XGBoost). All analyses were conducted in Python 3.9 using the \u003cem\u003escikit-learn\u003c/em\u003e package (v0.21.3).\u003c/p\u003e\u003cp\u003eBefore model training, the dataset was randomly divided into training, validation, and testing sets in a ratio of 6:2:2. Sample labels were numerically encoded using the LabelEncoder and to_categorical functions. To optimize model performance, hyperparameters were tuned using grid search (GridSearchCV) with five-fold cross-validation (cv\u0026thinsp;=\u0026thinsp;5). The parameter combination that achieved the highest average validation score was used to train the final model.\u003c/p\u003e\u003cp\u003eFor each algorithm, accuracy, precision, recall, F1-score, and five-fold cross-validation scores were calculated to assess performance stability. Receiver operating characteristic (ROC) curves and the corresponding area under the curve (AUC) were generated to visualize discriminative ability. Confusion matrices were constructed to compare the predicted and true categories, highlighting any misclassifications among ECC species.\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\n \u003ch2\u003e3.1 Strains identification\u003c/h2\u003e\n \u003cp\u003eUsing a 95% and 98% ANI cutoff to define species or subspecies boundaries, all 92 ECC strains were strictly identified. \u003cem\u003eE. hormaechei\u003c/em\u003e was the most common species detected (56/92, 60.9%), which can be subdivided into \u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003esteigerwaltii\u003c/em\u003e (28/92, 30.4%), \u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003exiangfangensis\u003c/em\u003e (17/92, 18.5%) and \u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003ehoffmannii\u003c/em\u003e (11/92, 12.0%). The remaining four species were \u003cem\u003eE. cloacae\u003c/em\u003e (12/92, 12.0%), \u003cem\u003eE. bugandensis\u003c/em\u003e (11/92, 12.0%), \u003cem\u003eE. kobei\u003c/em\u003e (7/92, 7.6%) and \u003cem\u003eE. roggenkampii\u003c/em\u003e (6/92, 6.5%). On the contary, the Vitek 2 compact indistinguishably identified all the strains as ECC, with a confidence value ranging from 95%-99%. The Vitek MS correctly identified one strain of \u003cem\u003eE. kobei\u003c/em\u003e with a confidence value of 99.9% while achieving ambiguous results for all the other isolates, the majority of which was misidentified as \u003cem\u003eE. cloacae\u003c/em\u003e and \u003cem\u003eE. asburiae\u003c/em\u003e, each with a 50.0% confidence value (Supplementary Table \u003cspan class=\"InternalRef\"\u003eS1\u003c/span\u003e).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\n \u003ch2\u003e3.2 Average, deconvoluted and characteristic peaks of SERS spectra\u003c/h2\u003e\n \u003cp\u003eThe full Raman spectra of bacteria contain morphological characteristics and physiological information of bacteria. In this study, the SERS spectra of ECC were collected separately. We computed the average Raman spectra and standard deviations of seven ECC species, thereby quantitatively revealing the overall trends and data variations among the SERS spectra of each bacterial type. As shown in Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.A-G, the SERS spectrum repeatability of the seven ECC species was good, and the reproducibility of the SERS spectra varying within an acceptable range. However, due to the morphological and physiological similarities among the bacterial cells of the seven ECC species, we also employed deconvolution techniques to generate SERS component bands directly associated with molecular structures. As can be seen from Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.H-N, the deconvoluted spectra are composed of a series of \u003cem\u003eVoigt\u003c/em\u003e sub-bands, where each sub-band represents a spectral characteristic peak. By deconstructing different sub-bands, the differences among the seven ECC species are amplified. This method effectively extracts the important characteristic peaks from the Raman spectra and eliminates interference from spurious peaks.\u003c/p\u003e\n \u003cp\u003eWe also examined the characteristic peaks of different ECC species based on SERS spectra (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.O-U). Due to the high similarities within the complex, there were still some identical peaks but with significantly varied intensities. For example, the molecular vibration at 658 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e represents COO- deformation of guanine [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e], appearing in both \u003cem\u003eE. hormaechei subsp. xiangfangensis\u003c/em\u003e and \u003cem\u003eE.cloacae\u003c/em\u003e, 730 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e represents adenine [\u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e], appearing in \u003cem\u003eE. hormaechei subsp. steigerwaltii\u003c/em\u003e, \u003cem\u003eE. hormaechei subsp. xiangfangensis\u003c/em\u003e, \u003cem\u003eE.cloacae\u003c/em\u003e, and \u003cem\u003eE. bugandensis\u003c/em\u003e. As for the unique characteristic peaks, they represented different molecular components and vibrations. For instance, in \u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003eSteigerwaltii\u003c/em\u003e, the peak at 1,329 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e was associated with Amide III band [\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e], 1,450 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e was associated with C\u0026ndash;H deformation [\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e], and 1,581 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e was assigned to the Ring breath Tyrosine [\u003cspan class=\"CitationRef\"\u003e21\u003c/span\u003e]. A prominent peak at 1,331 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e of \u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003ehoffmannii\u003c/em\u003e was associated with DNA vibration [\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e]. The vibration observed at 1,454 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e in \u003cem\u003eE.cloacae\u003c/em\u003e was attributed to CH2 stretching [\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e]. In \u003cem\u003eE. kobei\u003c/em\u003e, the unique peak at 661 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e and 1,332 cm-1 was assigned to glutathione and CAOAC str or ring breathing/guanine respectively [\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e\n \u003cp\u003eBased on the results above, \u003cem\u003eE. hormaechei subsp. steigerwaltii\u003c/em\u003e can be uniquely identified based on characteristic peaks at 1329 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e, 1450 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e or 1581 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e; \u003cem\u003eE. hormaechei subsp. hoffmannii\u003c/em\u003e be identified based on characteristic peaks at 731 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e, 1331 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e or 1579 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e; \u003cem\u003eE.cloacae\u003c/em\u003e be identified based on characteristic peaks at 1454 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e; \u003cem\u003eE. roggenkampii\u003c/em\u003e be identified based on characteristic peaks at 660 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e, 1004 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e, 1049 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e, 1333 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e or 1456 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e; \u003cem\u003eE. kobei\u003c/em\u003e be identified based on characteristic peaks at 661 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e, 1332 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e. \u003cem\u003eE. hormaechei subsp. xiangfangensis\u003c/em\u003e can be identified based on the characteristic peaks at 658 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e combined with 1330 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e or 1455 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e. \u003cem\u003eE. bugandensis\u003c/em\u003e can be identified based on the combination of characteristic peaks at 659 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e plus 1589 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003e\u003cstrong\u003e3.3 Clustering analysis using Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA)\u003c/strong\u003e\u003c/h2\u003e\n \u003cp\u003eClustering analysis was used to determine whether the seven different ECC species are separable based on their clustering in the feature coordinate system. We used the clustering method named OPLS-DA, to analyze the spectra of ECC. The results without normalization revealed that the OPLS-DA algorithm exhibit a relative lower degree of fitting between the features and spectral samples of input matrix (R2X\u0026thinsp;=\u0026thinsp;0.997, R2Y\u0026thinsp;=\u0026thinsp;0.666) and predictive capability for unknown samples (Q2\u0026thinsp;=\u0026thinsp;0.165) (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003eA). In contrast, OPLS-DA with normalization (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003eB) can better preserve the association rules between different spectra due to prior learning. This enables the model to exhibit a relative higher degree of fitting between the features and spectral samples of input matrix (R2X\u0026thinsp;=\u0026thinsp;0.980, R2Y\u0026thinsp;=\u0026thinsp;1.000), indicating that normalization can effectively distinguish the spectra of different ECC species. Furthmore, it also demonstrates predictive capability for unknown samples (Q2\u0026thinsp;=\u0026thinsp;0.313). However, overlapping of spectral data points still exists. Therefore, we need to seek more advanced machine learning methods to build rapid identification models of ECC.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003e3.4 Comparison of supervised learning algorithms\u003c/h2\u003e\n \u003cp\u003eTo enable quantitative prediction of ECC, we employed six machine learning algorithms to develop an optimal decision-making model. Before analyzing the SERS data, we present the results of each model\u0026apos;s parameter combinations obtained through grid search using a score gradient plot (Supplementary Table S2). We found that the recognition accuracy of each model was improved with the combination and iteration of parameters. Subsequently, the obtained best parameter combination was inputted into each function, and the models\u0026rsquo; performance were evaluated using five different evaluation metrics. As shown in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e, the XGBoost algorithm exhibited superior performance with the highest accuracy (accuracy\u0026thinsp;=\u0026thinsp;97.78%) and stability (5Fold\u0026thinsp;=\u0026thinsp;97.45%). Notably, Gboost, RF, SVM, and DT also achieved satisfactory results, possibly due to these algorithms belonging to ensemble learning, which possess the ability to handle non-linear relationships and strong feature selection. However, the AdaBoost algorithm failed to differentiate different species of ECC (accuracy\u0026thinsp;=\u0026thinsp;43.13%), because of the overlapping or intersecting SERS spectra of ECC in feature space, thereby affecting the feature selecting of AdaBoost.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"char\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison of the prediction capabilities of six supervised machine learning models in the analysis of the SERS spectra of \u003cem\u003eEnterobacter cloacae\u003c/em\u003e complexes.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAlgorithm\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eF1-score\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e5Fold\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eXGBoost\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e97.78%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e97.78%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e97.76%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e97.78%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e97.45%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eGBoost\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e97.49%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e97.49%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e97.44%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e97.49%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e96.67%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eRF\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e91.80%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e91.80%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e91.83%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e91.84%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e92.72%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eSVM\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e90.55%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e90.55%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e90.56%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e90.51%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e89.21%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eDT\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e84.56%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e84.56%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e84.54%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e84.55%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e82.10%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eAdaBoost\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e43.13%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e43.13%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e44.47%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e43.04%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e45.56%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003eROC curve and confusion matrix are commonly used to assess the performance of classification models. We employed the One-vs-All strategy to plot the ROC (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eA) curves for each model, evaluating their ability to discriminate false positive rate (FPR) and true positive rate (TPR) on the test set. The results demonstrate that XGBoost achieved the highest AUC value (AUC\u0026thinsp;=\u0026thinsp;0.9982), while the remaining algorithms also exhibited performance consistent with the metrics provided above. For the optimal classification model, we used the confusion matrix (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eB) to examine the model\u0026apos;s performance on spectral data in detail. It can be seen that XGBoost successfully identified all ECC spectra. For the performance on \u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003esteigerwaltii\u003c/em\u003e, 1.05% of the spectra were incorrectly classified as \u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003ehoffmanniic\u003c/em\u003e. There were also 0.51%, 0.51%, 0.51% of \u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003exiangfangensis\u003c/em\u003e spectra misidentifying as \u003cem\u003eE. cloacae\u003c/em\u003e, \u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003ehoffmanniic\u003c/em\u003e and \u003cem\u003eE\u003c/em\u003e, \u003cem\u003ebugandensis\u003c/em\u003e, respectively. The average recognition accuracy of the XGBoost model was 97.75%, which further demonstrated the potential of this algorithm in distinguishing SERS spectrum of different ECC.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eAccording to the China Antimicrobial Resistance Surveillance System (CARSS) data, \u003cem\u003eEnterobacter\u003c/em\u003e spp. rank fifth in the isolation rate of Gram-negative bacteria, accounting for 3\u0026ndash;5% of all bacterial isolates in China [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. \u003cem\u003eEnterobacter\u003c/em\u003e spp. isolated from clinical samples are usually reported as \u003cem\u003eE. cloacae\u003c/em\u003e, and sometimes \u003cem\u003eE. asburiae\u003c/em\u003e, \u003cem\u003eE. hormaechei\u003c/em\u003e, or \u003cem\u003eE. kobei\u003c/em\u003e by phenotypic methods, all belonging to ECC. Due to the similarities between species, the identification of ECC is often inaccurate and can be variable when repeating the test. However, the discrimination of ECC is clinically important due to the resistance pattern variations among different species. Researchers have shown that most of the carbapenem-resistant isolates were identified as \u003cem\u003eE. hormaechei subsp.xiangfangensis\u003c/em\u003e ST171, a clone circulating globally [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Besides, colistin hetero-resistance was found in all or most of \u003cem\u003eE. roggenkampii\u003c/em\u003e, \u003cem\u003eE. kobei\u003c/em\u003e, \u003cem\u003eE. chuandaensis\u003c/em\u003e and \u003cem\u003eE. cloacae\u003c/em\u003e but rarely seen in \u003cem\u003eE. hormaechei\u003c/em\u003e subspecies and \u003cem\u003eE. ludwigii\u003c/em\u003e [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Fortunately, with the development of new technologies, the taxonomy of ECC has evolved over time, from phenotypic methods [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], such as Gram staining, biochemical assays, and MALDI-TOF MS, to molecular approaches, such as 16S rRNA gene sequencing [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e], marker gene hsp60 sequencing [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] and WGS based ANI and DDH [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. The species and subspecies within ECC have been assigned to a more detailed classification. To date, twelve species including \u003cem\u003eE. cloacae\u003c/em\u003e, \u003cem\u003eE. hormaechei\u003c/em\u003e, \u003cem\u003eE. asburiae\u003c/em\u003e, \u003cem\u003eE. cancerogenus\u003c/em\u003e, \u003cem\u003eE. kobei\u003c/em\u003e, \u003cem\u003eE. ludwigii\u003c/em\u003e, \u003cem\u003eE. mori\u003c/em\u003e, \u003cem\u003eE. nimipressuralis\u003c/em\u003e, \u003cem\u003eE. roggenkampii\u003c/em\u003e, \u003cem\u003eE. chengduensis\u003c/em\u003e, and \u003cem\u003eE. bugandensis\u003c/em\u003e and \u003cem\u003eE. soli\u003c/em\u003e are assigned to ECC [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. \u003cem\u003eE. hormaechei\u003c/em\u003e can be subsequently divided into five subspecies (\u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003esteigerwaltii\u003c/em\u003e, subsp. \u003cem\u003eoharae\u003c/em\u003e, subsp. \u003cem\u003exiangfangensis\u003c/em\u003e, subsp. \u003cem\u003ehoffmannii\u003c/em\u003e, and subsp. \u003cem\u003ehormaechei\u003c/em\u003e), adding more complexities in the identification of ECC [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn clinical laboratories, the accurate identification of ECC species and subspecies still remains a challenge. Routine identification of ECC is mainly dependent on phenotypic characteristics by using commercialized systems, such as Vitek 2 compact and the MALDI-TOF MS technology [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e], despite that both methods can only give an ambiguous result. This was also confirmed in our study that Vitek 2 compact failed to assign ECC species and subspecies and Vitek MS only correctly identified one strain of \u003cem\u003eE. kobei\u003c/em\u003e. Molecular methods are more suitable for precisely identification of the ECC on species level. \u003cem\u003eHsp60\u003c/em\u003e typing is the earliest developed and currently most widely used method for this purpose [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. However, disadvantages are appearing due to the reclassification of species and subspecies, leading to unclassified or misclassified results by \u003cem\u003ehsp60\u003c/em\u003e. Recently, multi-plex real-time PCR and combination of single gene (\u003cem\u003ednaJ\u003c/em\u003e) real-time PCR plus MALDI-TOF MS for precise ECC identification have been reported [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. These were effective for limited species including \u003cem\u003eE. cloacae\u003c/em\u003e, \u003cem\u003eE. asbuiae\u003c/em\u003e, \u003cem\u003eE. hormaechei\u003c/em\u003e, \u003cem\u003eE. kobei\u003c/em\u003e and \u003cem\u003eE. ludwigii\u003c/em\u003e and unable to identify subspecies. So far, only WGS based ANI and DDH are reliable methods for accurate characterization of ECC species and subspecies. Considering the long-period and high-cost of WGS and the difficulty in data analysis, there is a significant requirement to seek easier and cheaper methods.\u003c/p\u003e\u003cp\u003eIn recent years, SERS coupled with machine learning algorithms has been emerging as a new technology for the rapid and accurate discrimination of various bacterial pathogens due to its strong Raman effects [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. Different SERS substrates and machine learning algorithms have been tried during the exploration of using SERS technique as a quantitative analytical tool for bacterial pathogen diagnosis. Though not developed long, SERS has been successfully applied in the identification of multiple bacteria, like \u003cem\u003eEscherichia coli\u003c/em\u003e, \u003cem\u003eSalmonella typhimurium\u003c/em\u003e, \u003cem\u003eStaphylococcus aureus\u003c/em\u003e, \u003cem\u003eStaphylococcus epidermidis\u003c/em\u003e, \u003cem\u003eBacillus megaterium\u003c/em\u003e and so on [\u003cspan additionalcitationids=\"CR39\" citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. Previous studies have demonstrated that simple average SERS spectral analysis plus machine learning algorithms is sufficient for discriminating biological samples based on significant variations in characteristic peaks [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. In this study we attempted for the first time to apply SERS in the identification of ECC. Since ECC consists of very much closely related species, it is challenging to accurately distinguish the SERS spectra of different species. To overcome this limitation, deconvoluted SERS spectra were generated to identify subtle molecular vibrations, which has been successfully used to detect differences in very similar species such as Candida and \u003cem\u003eShigella\u003c/em\u003e [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. On this basis, we extracted a list of characteristic peaks within ECC. Except for \u003cem\u003eE. hormaechei subsp. xiangfangensis\u003c/em\u003e and \u003cem\u003eE. bugandensis\u003c/em\u003e which needs to be identified based on characteristic peaks combination, the other five species all had at least one characteristic peak unique to themselves. However, despite being recognized as unique characteristic peaks, the molecular vibrations were quite close, posing high requirements in detection.\u003c/p\u003e\u003cp\u003eDespite our efforts to minimize undesired effects on Raman spectroscopy measurements during the acquisition of SERS signals, the measured spectral signals still encompass extraneous contributions from the instrument or the sample itself. Hence, data cleaning becomes imperative to eliminate these detrimental effects [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. Thus, we processed the spectral feature matrix with maximum and minimum normalization and baseline correction. Subsequently, the processed feature matrices were used for OPLS-DA clustering analyses. This algorithm used specific shapes and combined features to determine sample clusters. The R2X, R2Y, and Q2 metrics were used to appraise ECC species and assess the quality of SERS data. Neverthelss, the high dimensionality and similarity of the SERS spectral data still pose significant challenges for the OPLS-DA clustering algorithm to effectively distinguish different ECC species. Therefore, exploring more advanced methods for spectral data analysis is necessary.\u003c/p\u003e\u003cp\u003eCiloglu FU et al combined SERS and deep learning techniques for drug-resistant \u003cem\u003eStaphylococcus aureus\u003c/em\u003e detection, achieving an accuracy of 97.66% and an AUC of 0.99 [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. This breakthrough garnered significant attention and spurred numerous researchers to explore intelligent spectral analysis. In this study, we used normalized and baseline-corrected spectral data as input to construct six ensemble learning models. We employed the GridSearch algorithm to analyze the appropriate hyperparameters of various models [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. The parameter combination yielding the highest final score for each model was selected to examine SERS data and identify different species of ECC in an independent dataset. Among the six models, XGBoost demonstrated the most accurate diagnosis with the highest efficiency in analyzing different species of ECC. This novel diagnostic method holds a promising potential to attain early prediction of ECC, surpassing the capabilities of existing methods.\u003c/p\u003e\u003cp\u003eIn summary, SERS coupled with machine learning algorithms showed the potential in the identification of highly similar ECC, enabling us to understand the aspect of clinical significance, epidemiology, and drug resistance of ECC at the species level. In comparison to currently available phenotypic and molecular methods, this method is rapid, accurate, and cost-efficient that is suitable for future use in routine diagnostic tests.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e\n\u003cp\u003eThis study was reviewed and approved by the Ethics Committee of Peking Union Medical College Hospital (Approval No. I-22PJ396). Only identified ECC isolates collected during routine clinical diagnostics were used, and no patient-related information was accessed. Therefore, the requirement for informed consent was waived by the Ethics Committee in accordance with the \u0026ldquo;Ethical Review Measures for Biomedical Research Involving Humans\u0026rdquo; (National Health and Family Planning Commission of China, 2016) and the \u0026ldquo;Regulations of the People\u0026rsquo;s Republic of China on the Administration of Human Genetic Resources\u0026rdquo; (State Council Order No. 717, 2019). This study complied with the Declaration of Helsinki.\u003c/p\u003e\n\u003ch2\u003eClinical trial\u003c/h2\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003ch2\u003eConsent for publication\u003c/h2\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003ch2\u003eCompeting interests\u003c/h2\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003ch2\u003eFundings\u003c/h2\u003e\n\u003cp\u003eThis work was supported by National Science Foundation for Young Scientists of China (82202541), Peking Union Medical College Hospital Talent Cultivation Program Category D (UHB12396) and Fundamental Research Funds for the Central Universities (3332022012).\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eMZ, XX, and YL contributed equally to this work. MZ and XX designed and performed the experiments, collected the data, and carried out the machine learning analysis. YL contributed to data interpretation, microbiological identification, and manuscript preparation. YX supervised the microbiological methodology and provided critical revision of the manuscript. BG and JG conceived and supervised the study, provided resources and finalized the manuscript. All authors read and approved the final version of the manuscript.\u003c/p\u003e\n\u003ch2\u003eAcknowledgements\u003c/h2\u003e\n\u003cp\u003eNone.\u003c/p\u003e\n\u003ch2\u003eData Availability\u003c/h2\u003e\n\u003cp\u003eAll whole-genome sequencing data from this study have been deposited in the GenBank under BioProject accession No. PRJNA1226973.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eMiller WR, Arias CA. ESKAPE pathogens: antimicrobial resistance, epidemiology, clinical impact and therapeutics. Nat Rev Microbiol. 2024;22(10):598\u0026ndash;616. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41579-024-01054-w\u003c/span\u003e\u003cspan address=\"10.1038/s41579-024-01054-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMezzatesta ML, Gona F, Stefani S. Enterobacter cloacae complex: clinical impact and emerging antibiotic resistance. Future Microbiol. 2012;7(7):887\u0026ndash;902. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.2217/fmb.12.61\u003c/span\u003e\u003cspan address=\"10.2217/fmb.12.61\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTindall BJ, Sutton G, Garrity GM. Enterobacter aerogenes Hormaeche and Edwards 1960 (Approved Lists 1980) and Klebsiella mobilis, Bascomb et al. 1971 (Approved Lists 1980) share the same nomenclatural type (ATCC 13048) on the Approved Lists and are homotypic synonyms, with consequences for the name Klebsiella mobilis Bascomb 1971 (Approved Lists 1980). Int J Syst Evol Microbiol. 2017;67(2):502-4; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1099/ijsem.0.001572\u003c/span\u003e\u003cspan address=\"10.1099/ijsem.0.001572\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGavini F, Mergaert J, Beji A, Mielcarek C, Izard D, Kersters K, et al. Transfer of Enterobacter agglomerans (Beijerinck 1888) Ewing and Fife 1972 to Pantoea gen. nov. as Pantoea agglomerans comb, nov. and Description of Pantoea dispersa sp. nov. Int J Syst Evol MicroBiol. 1989;39(3):337\u0026ndash;45. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1099/00207713-39-3-337\u003c/span\u003e\u003cspan address=\"10.1099/00207713-39-3-337\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIversen C, Mullane N, McCardell B, Tall BD, Lehner A, Fanning S, et al. Cronobacter gen. nov., a new genus to accommodate the biogroups of Enterobacter sakazakii, and proposal of Cronobacter sakazakii gen. nov., comb. nov., Cronobacter malonaticus sp. nov., Cronobacter turicensis sp. nov., Cronobacter muytjensii sp. nov., Cronobacter dublinensis sp. nov., Cronobacter genomospecies 1, and of three subspecies, Cronobacter dublinensis subsp. dublinensis subsp. nov., Cronobacter dublinensis subsp. lausannensis subsp. nov. and Cronobacter dublinensis subsp. lactaridi subsp. nov. Int J Syst Evol MicroBiol. 2008;58(6):1442\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1099/ijs.0.65577-0\u003c/span\u003e\u003cspan address=\"10.1099/ijs.0.65577-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWisplinghoff H, Bischoff T, Tallent SM, Seifert H, Wenzel RP, Edmond MB. Nosocomial bloodstream infections in US hospitals: analysis of 24,179 cases from a prospective nationwide surveillance study. Clin Infect Dis. 2004;39(3):309\u0026ndash;17. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1086/421946\u003c/span\u003e\u003cspan address=\"10.1086/421946\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhou K, Yu W, Cao X, Shen P, Lu H, Luo Q, et al. Characterization of the population structure, drug resistance mechanisms and plasmids of the community-associated Enterobacter cloacae complex in China. J Antimicrob Chemother. 2018;73(1):66\u0026ndash;76. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/jac/dkx361\u003c/span\u003e\u003cspan address=\"10.1093/jac/dkx361\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZong Z, Feng Y, McNally A. Carbapenem and Colistin Resistance in Enterobacter: Determinants and Clones. Trends Microbiol. 2021;29(6):473\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.tim.2020.12.009\u003c/span\u003e\u003cspan address=\"10.1016/j.tim.2020.12.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGiamarellou H. Epidemiology of infections caused by polymyxin-resistant pathogens. Int J Antimicrob Agents. 2016;48(6):614\u0026ndash;21. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijantimicag.2016.09.025\u003c/span\u003e\u003cspan address=\"10.1016/j.ijantimicag.2016.09.025\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFukuzawa S, Sato T, Aoki K, Yamamoto S, Ogasawara N, Nakajima C, et al. High prevalence of colistin heteroresistance in specific species and lineages of Enterobacter cloacae complex derived from human clinical specimens. Ann Clin Microbiol Antimicrob. 2023;22(1):60. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12941-023-00610-1\u003c/span\u003e\u003cspan address=\"10.1186/s12941-023-00610-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGuerin F, Isnard C, Sinel C, Morand P, Dhalluin A, Cattoir V, et al. Cluster-dependent colistin hetero-resistance in Enterobacter cloacae complex. J Antimicrob Chemother. 2016;71(11):3058\u0026ndash;61. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/jac/dkw260\u003c/span\u003e\u003cspan address=\"10.1093/jac/dkw260\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTang JW, Li JQ, Yin XC, Xu WW, Pan YC, Liu QH, et al. Rapid Discrimination of Clinically Important Pathogens Through Machine Learning Analysis of Surface Enhanced Raman Spectra. Front Microbiol. 2022;13:843417. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fmicb.2022.843417\u003c/span\u003e\u003cspan address=\"10.3389/fmicb.2022.843417\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu W, Tang JW, Mou JY, Lyu JW, Di YW, Liao YL, et al. Rapid discrimination of Shigella spp. and Escherichia coli via label-free surface enhanced Raman spectroscopy coupled with machine learning algorithms. Front Microbiol. 2023;14:1101357. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fmicb.2023.1101357\u003c/span\u003e\u003cspan address=\"10.3389/fmicb.2023.1101357\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXiong XS, Yao LF, Luo YF, Yuan Q, Si YT, Chen J, et al. Differentiation of closely-related species within Acinetobacter baumannii-calcoaceticus complex via Raman spectroscopy: a comparative machine learning analysis. World J Microbiol Biotechnol. 2024;40(5):146. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11274-024-03948-6\u003c/span\u003e\u003cspan address=\"10.1007/s11274-024-03948-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKonstantinidis KT, Tiedje JM. Genomic insights that advance the species definition for prokaryotes. Proc Natl Acad Sci U S A. 2005;102(7):2567\u0026ndash;72. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1073/pnas.0409727102\u003c/span\u003e\u003cspan address=\"10.1073/pnas.0409727102\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTang JW, Li F, Liu X, Wang JT, Xiong XS, Lu XY, et al. Detection of Helicobacter pylori infection in human gastric fluid through surface-enhanced Raman spectroscopy coupled with machine learning algorithms. Lab Invest. 2023;100310. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.labinv.2023.100310\u003c/span\u003e\u003cspan address=\"10.1016/j.labinv.2023.100310\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCiloglu FU, Caliskan A, Saridag AM, Kilic IH, Tokmakci M, Kahraman M, et al. Drug-resistant Staphylococcus aureus bacteria detection by combining surface-enhanced Raman spectroscopy (SERS) and deep learning techniques. Sci Rep. 2021;11(1):18444. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-021-97882-4\u003c/span\u003e\u003cspan address=\"10.1038/s41598-021-97882-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen X, Tang M, Liu Y, Huang J, Liu Z, Tian H, et al. Surface-enhanced Raman scattering method for the identification of methicillin-resistant Staphylococcus aureus using positively charged silver nanoparticles. Mikrochim Acta. 2019;186(2):102. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00604-018-3150-6\u003c/span\u003e\u003cspan address=\"10.1007/s00604-018-3150-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang X, Ren L, Diao Z, He Y, Zhang J, Liu M, et al. Robust Spontaneous Raman Flow Cytometry for Single-Cell Metabolic Phenome Profiling via pDEP-DLD-RFC. Adv Sci (Weinh). 2023;10(16):e2207497. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/advs.202207497\u003c/span\u003e\u003cspan address=\"10.1002/advs.202207497\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFan C, Hu Z, Mustapha A, Lin M. Rapid detection of food- and waterborne bacteria using surface-enhanced Raman spectroscopy coupled with silver nanosubstrates. Appl Microbiol Biotechnol. 2011;92(5):1053\u0026ndash;61. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00253-011-3634-3\u003c/span\u003e\u003cspan address=\"10.1007/s00253-011-3634-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYaacob MF, Murata A, Nor NHM, Jesse FFA, Raja Yahya MFZ. Biochemical composition, morphology and antimicrobial susceptibility pattern of Corynebacterium pseudotuberculosis biofilm. J King Saud Univ - Sci. 2021;33(1):101225. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jksus.2020.10.022\u003c/span\u003e\u003cspan address=\"10.1016/j.jksus.2020.10.022\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWalter A, Marz A, Schumacher W, Rosch P, Popp J. Towards a fast, high specific and reliable discrimination of bacteria on strain level by means of SERS in a microfluidic device. Lab Chip. 2011;11(6):1013\u0026ndash;21. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1039/c0lc00536c\u003c/span\u003e\u003cspan address=\"10.1039/c0lc00536c\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLu X, Rasco BA, Jabal JM, Aston DE, Lin M, Konkel ME. Investigating antibacterial effects of garlic (Allium sativum) concentrate and garlic-derived organosulfur compounds on Campylobacter jejuni by using Fourier transform infrared spectroscopy, Raman spectroscopy, and electron microscopy. Appl Environ Microbiol. 2011;77(15):5257\u0026ndash;69. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1128/AEM.02845-10\u003c/span\u003e\u003cspan address=\"10.1128/AEM.02845-10\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDu Y, Han D, Liu S, Sun X, Ning B, Han T, et al. Raman spectroscopy-based adversarial network combined with SVM for detection of foodborne pathogenic bacteria. Talanta. 2022;237:122901. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.talanta.2021.122901\u003c/span\u003e\u003cspan address=\"10.1016/j.talanta.2021.122901\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTorre-Gutierrez LG, Martinez-Zerega BE, Oseguera-Galindo DO, Aguilar-Lemarroy A, Jave-Suarez LF, Torres-Gonzalez LA, et al. Breast cancer chemotherapy treatment monitoring based on serum sample Raman spectroscopy. Lasers Med Sci. 2022;37(9):3649\u0026ndash;59. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10103-022-03646-5\u003c/span\u003e\u003cspan address=\"10.1007/s10103-022-03646-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eQin X, Ding L, Hao M, Li P, Hu F, Wang M. Antimicrobial resistance of clinical bacterial isolates in China: current status and trends. JAC Antimicrob Resist. 2024;6(2):dlae052. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/jacamr/dlae052\u003c/span\u003e\u003cspan address=\"10.1093/jacamr/dlae052\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGanbold M, Seo J, Wi YM, Kwon KT, Ko KS. Species identification, antibiotic resistance, and virulence in Enterobacter cloacae complex clinical isolates from South Korea. Front Microbiol. 2023;14:1122691. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fmicb.2023.1122691\u003c/span\u003e\u003cspan address=\"10.3389/fmicb.2023.1122691\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGeorgiades K, Raoult D. Defining pathogenic bacterial species in the genomic era. Front Microbiol. 2010;1:151. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fmicb.2010.00151\u003c/span\u003e\u003cspan address=\"10.3389/fmicb.2010.00151\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim M, Oh HS, Park SC, Chun J. Towards a taxonomic coherence between average nucleotide identity and 16S rRNA gene sequence similarity for species demarcation of prokaryotes. Int J Syst Evol Microbiol. 2014;64(Pt 2):346\u0026ndash;51. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1099/ijs.0.059774-0\u003c/span\u003e\u003cspan address=\"10.1099/ijs.0.059774-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHoffmann H, Roggenkamp A. Population genetics of the nomenspecies Enterobacter cloacae. Appl Environ Microbiol. 2003;69(9):5306\u0026ndash;18. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1128/aem.69.9.5306-5318.2003\u003c/span\u003e\u003cspan address=\"10.1128/aem.69.9.5306-5318.2003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRichter M, Rossello-Mora R. Shifting the genomic gold standard for the prokaryotic species definition. Proc Natl Acad Sci U S A. 2009;106(45):19126\u0026ndash;31. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1073/pnas.0906412106\u003c/span\u003e\u003cspan address=\"10.1073/pnas.0906412106\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGoris J, Konstantinidis KT, Klappenbach JA, Coenye T, Vandamme P, Tiedje JM. DNA-DNA hybridization values and their relationship to whole-genome sequence similarities. Int J Syst Evol Microbiol. 2007;57(Pt 1):81\u0026ndash;91. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1099/ijs.0.64483-0\u003c/span\u003e\u003cspan address=\"10.1099/ijs.0.64483-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eQiu X, Ye K, Ma Y, Zhao Q, Wang L, Yang J. Genome sequence-based species classification of Enterobacter cloacae complex: a study among clinical isolates. Microbiol Spectr. 2024;12(6):e0431223. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1128/spectrum.04312-23\u003c/span\u003e\u003cspan address=\"10.1128/spectrum.04312-23\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDe Florio L, Riva E, Giona A, Dedej E, Fogolari M, Cella E, et al. MALDI-TOF MS Identification and Clustering Applied to Enterobacter Species in Nosocomial Setting. Front Microbiol. 2018;9:1885. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fmicb.2018.01885\u003c/span\u003e\u003cspan address=\"10.3389/fmicb.2018.01885\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJi Y, Wang P, Xu T, Zhou Y, Chen R, Zhu H, et al. Development of a One-Step Multiplex PCR Assay for Differential Detection of Four species (Enterobacter cloacae, Enterobacter hormaechei, Enterobacter roggenkampii, and Enterobacter kobei) Belonging to Enterobacter cloacae Complex With Clinical Significance. Front Cell Infect Microbiol. 2021;11:677089. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fcimb.2021.677089\u003c/span\u003e\u003cspan address=\"10.3389/fcimb.2021.677089\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePavlovic M, Konrad R, Iwobi AN, Sing A, Busch U, Huber I. A dual approach employing MALDI-TOF MS and real-time PCR for fast species identification within the Enterobacter cloacae complex. FEMS Microbiol Lett. 2012;328(1):46\u0026ndash;53. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/j.1574-6968.2011.02479.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1574-6968.2011.02479.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang L, Liu W, Tang JW, Wang JJ, Liu QH, Wen PB, et al. Applications of Raman Spectroscopy in Bacterial Infections: Principles, Advantages, and Shortcomings. Front Microbiol. 2021;12:683580. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fmicb.2021.683580\u003c/span\u003e\u003cspan address=\"10.3389/fmicb.2021.683580\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang Y, Lee K, Irudayaraj J. Silver Nanosphere SERS Probes for Sensitive Identification of Pathogens. J Phys Chem C. 2010;114(39):16122\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1021/jp1015406\u003c/span\u003e\u003cspan address=\"10.1021/jp1015406\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRodriguez L, Zhang Z, Wang D. Recent advances of Raman spectroscopy for the analysis of bacteria. Anal Sci Adv. 2023;4(3\u0026ndash;4):81\u0026ndash;95. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/ansa.202200066\u003c/span\u003e\u003cspan address=\"10.1002/ansa.202200066\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhou S, Lu C, Li Y, Xue L, Zhao C, Tian G, et al. Gold Nanobones Enhanced Ultrasensitive Surface-Enhanced Raman Scattering Aptasensor for Detecting Escherichia coli O157:H7. ACS Sens. 2020;5(2):588\u0026ndash;96. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1021/acssensors.9b02600\u003c/span\u003e\u003cspan address=\"10.1021/acssensors.9b02600\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang L, Tang JW, Li F, Usman M, Wu CY, Liu QH, et al. Identification of Bacterial Pathogens at Genus and Species Levels through Combination of Raman Spectrometry and Deep-Learning Algorithms. Microbiol Spectr. 2022;10(6):e0258022. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1128/spectrum.02580-22\u003c/span\u003e\u003cspan address=\"10.1128/spectrum.02580-22\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang N, Wang Y, Wang X, Zhang F, Xiao Y, Yan B, et al. Label-Free Detection of DNA Supramolecular Structure Formation by Surface-Enhanced Raman Spectroscopy. J Phys Chem Lett. 2022;13(26):6208\u0026ndash;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1021/acs.jpclett.2c01461\u003c/span\u003e\u003cspan address=\"10.1021/acs.jpclett.2c01461\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePezzotti G, Kobara M, Nakaya T, Imamura H, Fujii T, Miyamoto N, et al. Raman Metabolomics of Candida auris Clades: Profiling and Barcode Identification. Int J Mol Sci. 2022;23(19). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/ijms231911736\u003c/span\u003e\u003cspan address=\"10.3390/ijms231911736\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGuo S, Popp J, Bocklitz T. Chemometric analysis in Raman spectroscopy from experimental design to machine learning-based modeling. Nat Protoc. 2021;16(12):5426\u0026ndash;59. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41596-021-00620-3\u003c/span\u003e\u003cspan address=\"10.1038/s41596-021-00620-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMa ZW, Tang JW, Liu QH, Mou JY, Qiao R, Du Y, et al. Identification of geographic origins of Morus alba Linn. through surfaced enhanced Raman spectrometry and machine learning algorithms. J Biomol Struct Dyn. 2023;41(23):14285\u0026ndash;98. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1080/07391102.2023.2180433\u003c/span\u003e\u003cspan address=\"10.1080/07391102.2023.2180433\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTang JW, Qiao R, Xiong XS, Tang BX, He YW, Yang YY, et al. Rapid discrimination of glycogen particles originated from different eukaryotic organisms. Int J Biol Macromol. 2022;222(Pt A):1027\u0026ndash;36. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijbiomac.2022.09.233\u003c/span\u003e\u003cspan address=\"10.1016/j.ijbiomac.2022.09.233\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-microbiology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"mcro","sideBox":"Learn more about [BMC Microbiology](http://bmcmicrobiol.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/mcro","title":"BMC Microbiology","twitterHandle":"#bmcmicrobiology","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"surface enhanced Raman spectra, machine learning, Enterobacter cloacae complex, rapid identification","lastPublishedDoi":"10.21203/rs.3.rs-7939547/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7939547/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003e\u003cem\u003eEnterobacter cloacae\u003c/em\u003e complex (ECC) is an important nosocomial pathogen and consists of multiple similar species. The taxonomy of ECC has been consecutively updated, adding to its identification difficulty.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eA total of 92 ECC strains isolated from bloodstream infections during 2015\u0026ndash;2020 were collected from a tertiary hospital in China. All the strains were identified by Vitek 2 Compact and Vitek MS and then subjected to whole genome sequencing (WGS) for average nucleotide identity (ANI) analysis. Surfaced-enhanced Raman spectroscopy (SERS) with a set of machine learning algorithms was applied in identifying species within ECC.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eSeven species were identified through ANI, including 28 \u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003esteigerwaltii\u003c/em\u003e, 17 \u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003exiangfangensis\u003c/em\u003e, 12 \u003cem\u003eE. cloacae\u003c/em\u003e, 11 each of \u003cem\u003eE. hormaechei\u003c/em\u003e subsp. \u003cem\u003ehoffmannii\u003c/em\u003e and \u003cem\u003eE. bugandensis\u003c/em\u003e, seven \u003cem\u003eE. kobei\u003c/em\u003e and six \u003cem\u003eE. roggenkampii\u003c/em\u003e. The Vitek 2 compact indistinguishably identified all the strains as ECC and Vitek MS correctly identified one strain of \u003cem\u003eE. kobei\u003c/em\u003e while achieving ambiguous results for all the other isolates. SERS combined with XGBoost model achieved 97.75% accuracy with an area under the ROC curve value of 0.9982 in the identification of ECC.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eSERS coupled with machine learning algorithms holds a promising potential to acquire early prediction of ECC, outperforming the capabilities of other methods.\u003c/p\u003e","manuscriptTitle":"Development of Surface Enhanced Raman Spectra coupled with Machine Learning Analysis for Differentiation of Closely Related Species within Enterobacter cloacae Complex","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-13 16:05:08","doi":"10.21203/rs.3.rs-7939547/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-12-15T09:18:15+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-08T15:37:39+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-29T21:38:56+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"259648020034292641774426462685304233496","date":"2025-11-24T06:58:20+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"294003896981642751474584355575436219556","date":"2025-11-21T09:38:51+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"22383277690564705140794315778189380605","date":"2025-11-12T05:52:08+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-11-04T03:42:06+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-11-02T09:56:15+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-10-29T04:58:19+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-28T09:28:12+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Microbiology","date":"2025-10-28T09:24:54+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-microbiology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"mcro","sideBox":"Learn more about [BMC Microbiology](http://bmcmicrobiol.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/mcro","title":"BMC Microbiology","twitterHandle":"#bmcmicrobiology","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"b66b8d72-1df8-4782-9918-71727acf9ffb","owner":[],"postedDate":"November 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-02-09T16:15:38+00:00","versionOfRecord":{"articleIdentity":"rs-7939547","link":"https://doi.org/10.1186/s12866-026-04769-3","journal":{"identity":"bmc-microbiology","isVorOnly":false,"title":"BMC Microbiology"},"publishedOn":"2026-02-07 15:58:53","publishedOnDateReadable":"February 7th, 2026"},"versionCreatedAt":"2025-11-13 16:05:08","video":"","vorDoi":"10.1186/s12866-026-04769-3","vorDoiUrl":"https://doi.org/10.1186/s12866-026-04769-3","workflowStages":[]},"version":"v1","identity":"rs-7939547","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7939547","identity":"rs-7939547","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}