A method for detecting antibiotic resistance of escherichia coli based on sers and machine learning

Abstract

The present application relates to the technical field of escherichia coli drug resistance analysis, and particularly relates to a detection method for escherichia coli antibiotic resistance based on SERS and machine learning. The present application provides a detection method for escherichia coli antibiotic resistance based on SERS and machine learning to overcome the defects that high-precision and rapid diagnosis cannot be achieved in the prior art. The present application aims to utilize the core-shell nanostructure SERS substrate of silver-coated gold nanoparticles, combine the SERS spectrum differences of different phenotypic drug-resistant escherichia coli, and realize rapid and high-precision drug-resistant bacteria identification through machine learning algorithms. Through this method, the detection time can be significantly shortened, the detection accuracy can be improved, and accurate drug treatment plans can be provided for the clinic, which has a wide application prospect.

Description

Technical Field

[0001] This invention relates to the field of Escherichia coli antibiotic resistance analysis technology, specifically to a method for detecting Escherichia coli antibiotic resistance based on SERS and machine learning. Background Technology

[0002] The widespread use and overuse of antibiotics has led to the rapid development of drug-resistant bacteria (superbugs), which has become a major global public health problem. The emergence of drug-resistant bacteria renders conventional antibiotic treatments ineffective, increasing the difficulty and risk of treating clinical infectious diseases. Therefore, rapid and accurate detection and identification of drug-resistant bacteria are crucial for guiding clinical medication and improving treatment efficiency. Currently, traditional methods for detecting drug-resistant bacteria mainly include drug susceptibility testing, gene detection, and mass spectrometry analysis.

[0003] Traditional drug susceptibility tests (such as disk diffusion and broth dilution methods) are widely used in clinical drug resistance testing. While these methods can provide relatively reliable results, they are complex and time-consuming, typically requiring 24 to 48 hours or even longer. Furthermore, because these methods are based on bacterial growth, they require time to culture the bacteria, making them unsuitable for the urgent clinical need for rapid diagnosis, especially in the treatment of critically ill patients.

[0004] Molecular biology methods (PCR testing) can rapidly identify drug-resistant strains by detecting gene mutations associated with drug resistance. However, these techniques require sophisticated experimental equipment and skilled personnel, are costly, and cannot reflect the actual drug response of bacteria. Instead, they are based on specific genetic markers and are prone to false positive or false negative results.

[0005] Mass spectrometry analysis (such as MALDI-TOF MS) has made some progress in clinical microbiology identification in recent years, enabling bacterial identification and partial drug resistance analysis by analyzing bacterial protein or metabolite spectra. However, this technology has high requirements for sample processing and also suffers from problems such as long detection time and expensive equipment.

[0006] Faced with these challenges, surface-enhanced Raman scattering (SERS) technology has gradually become a research hotspot in the field of microbial detection. SERS is a spectroscopic analysis technique based on the localized plasmon resonance effect on the surface of nanomaterials, which can significantly enhance the Raman signal of target molecules, thus showing great potential for ultrasensitive detection. The advantages of SERS lie in its high sensitivity, rapid detection capability, and ability to analyze a variety of compounds in complex environments. Meanwhile, the rapid development of machine learning technology has provided a powerful tool for the analysis of massive amounts of data. Machine learning can analyze and recognize patterns in complex spectral data. Combining SERS with machine learning can extract useful information from complex SERS spectral data, automatically identifying and classifying bacteria.

[0007] Currently, there are many detection methods that combine SERS with machine learning, but they all suffer from inaccurate identification, inability to detect quickly, inability to accurately determine the drug resistance of E. coli, and most of them are based on a single simulated sample, which cannot be applied to real samples. Summary of the Invention

[0008] This invention aims to overcome the limitations of existing technologies in achieving high-precision and rapid diagnosis by providing a method for detecting antibiotic resistance in *E. coli* based on SERS and machine learning. This invention utilizes a core-shell nanostructured SERS substrate of silver-coated gold nanoparticles, combined with the SERS spectral differences of drug-resistant *E. coli* with different phenotypes, and employs machine learning algorithms to achieve rapid and high-precision identification of drug-resistant bacteria. This method can significantly shorten detection time, improve detection accuracy, and assist in providing precise drug treatment plans, demonstrating broad application prospects.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] This invention provides a method for detecting antibiotic resistance in Escherichia coli based on SERS and machine learning, comprising:

[0011] S1.1 Obtain the SERS spectra of drug-resistant Escherichia coli and normalize them to obtain SERS spectral data; obtain the resistance gene results and virulence gene results of drug-resistant Escherichia coli and binarize the gene expression to obtain resistance gene data and virulence gene data respectively;

[0012] S1.2 Input the SERS spectral data from S1.1 into the CNN model, and input the resistance gene data and virulence gene data from S1.1 into the MLP model; use tagged E. coli drug resistance data to train and validate the CNN model and the MLP model, and fuse the CNN model and the MLP model to obtain the E. coli bacterial drug resistance analysis simulation model.

[0013] S1.3 Obtain the SERS spectrum of the drug-resistant Escherichia coli to be tested, and normalize it to obtain SERS spectral data; obtain the resistance gene results and virulence gene results of the drug-resistant Escherichia coli to be tested, and binarize the gene expression to obtain resistance gene data and virulence gene data respectively;

[0014] S1.4 Input the SERS spectral data, resistance gene data and virulence gene data obtained in S1.3 into the Escherichia coli bacterial resistance analysis simulation model to analyze and obtain the drug resistance results of the target Escherichia coli.

[0015] Genotype and phenotype are two distinct dimensions describing bacterial drug resistance. Genotype data alone cannot fully reflect phenotypic resistance, thus failing to effectively guide antibiotic application. Bacterial phenotypic resistance depends not only on genetic factors but also on bacterial protein expression and metabolites. SERS technology provides information on bacterial surface proteins and metabolites. Therefore, this invention combines genotype and spectral data to train the model, significantly improving the accuracy of the E. coli drug resistance analysis simulation model in detecting E. coli drug resistance.

[0016] Specifically, by utilizing a specific SERS substrate material—silver-coated gold nanoparticles—the Raman signals of specific molecules on the surface of *E. coli* can be significantly amplified through their highly enhanced signal-capturing capabilities, thus generating clear SERS spectra. SERS spectroscopy reveals metabolites, surface proteins, and some genetic characteristics of *E. coli*, which are then combined with resistance gene and virulence gene results. Using a database of *E. coli* with known drug resistance results as tags, the model is trained to establish relationships between SERS spectra, resistance gene results, virulence gene results, and *E. coli* phenotypic drug resistance. Therefore, the successfully trained simulation model for analyzing *E. coli* bacterial drug resistance can obtain the drug resistance results of a target *E. coli* simply by inputting its SERS spectral results, resistance gene results, and virulence gene results, without requiring complex drug susceptibility testing. This significantly reduces detection time costs and enables rapid detection and interpretation of drug resistance results.

[0017] Therefore, the detection method established in this invention utilizes a specific substrate material to respond to high-precision SERS spectra, and combines SERS spectra, resistance genes, and virulence genes as the training dataset for the model. Using a database of E. coli with known drug resistance results as labels, the model is trained to obtain the relationship between SERS spectra, resistance genes, virulence genes, and E. coli phenotypic drug resistance. Ultimately, the complex and time-consuming drug susceptibility testing is transformed into a shorter SERS technique and PCR test, resulting in a more accurate and faster method for detecting E. coli drug resistance.

[0018] Preferably, in S1.1 and S1.3, the SERS spectra are normalized after baseline correction, and each SERS spectrum is measured at least 20 times.

[0019] SERS spectra are preprocessed to improve spectral quality and accuracy. Furthermore, each spectrum is measured multiple times to ensure repeatability and accuracy, resulting in the final SERS spectrum.

[0020] Preferably, in S1.1 and S1.3, the SERS spectrum is obtained by SERS technology after mixing silver-coated gold nanoparticles with drug-resistant Escherichia coli.

[0021] Preferably, the test parameters for the SERS spectrum are in the range of 600~1800 cm⁻¹. -1 Integration time is 2-3 seconds, laser wavelength is 785 nm, and resolution is 3-4 cm. -1 Laser power 350~380 mW, laser spot size 100~120 μm, cumulative time 5 times.

[0022] Preferably, the preparation method of the silver-coated gold nanoparticles is as follows:

[0023] Under heating conditions S5.1, chloroauric acid solution and sodium citrate solution are mixed. After the solution changes from yellow to dark red, it is stirred for 10-20 min and then cooled to obtain a gold nanoparticle suspension.

[0024] S5.2 Add ascorbic acid and silver nitrate sequentially to the gold nanoparticle suspension. After the suspension color changes from purple-red to orange-yellow, shake for 20-30 min to obtain a silver-coated gold nanoparticle solution.

[0025] Preferably, in S1.2, the labeled Escherichia coli drug resistance data is Escherichia coli phenotypic drug resistance data obtained from drug susceptibility testing.

[0026] Preferably, in S1.1, the drug-resistant Escherichia coli is isolated from the biological sample; in S1.3, the drug-resistant Escherichia coli to be tested is isolated from the biological sample.

[0027] Preferably, the biological sample is selected from blood, urine, sputum, or wound secretions.

[0028] Real-world samples, such as blood, urine, sputum, or wound secretions, come from the clinical samples of different patients and contain strains with various drug resistance mechanisms, many of which are highly resistant. Therefore, using these real-world samples as data for the model ensures its broad applicability.

[0029] Preferably, in S1.1 and S1.3, the results of resistance genes and virulence genes of drug-resistant Escherichia coli are obtained by sequencing technology.

[0030] Preferably, in S1.2, during the fusion process, the resistance gene data and virulence gene data are fused with the SERS spectral data respectively. During the fusion, the data in the resistance gene data and virulence gene data that have poor correlation with the SERS spectral data are cleaned and removed, and the remaining data continue to be fused with the SERS spectral data.

[0031] Therefore, the present invention has the following beneficial effects:

[0032] (1) This invention utilizes SERS technology combined with machine learning methods to complete the detection within a few hours (2.5 hours), which greatly shortens the time from sample collection to result output.

[0033] (2) This invention utilizes a specific SERS substrate material to amplify the SERS spectral signal, thereby more accurately reflecting the spectral characteristic parameters and capturing important characteristic signals. The simulation model for analyzing the drug resistance of Escherichia coli obtained by training has higher precision.

[0034] (3) This invention utilizes specific CNN and MLP models for machine learning. The phenotypic drug resistance data obtained from drug sensitivity experiments are used as labels. The SERS spectrum results, resistance gene results, and virulence gene results are input. The model is trained to obtain the relationship between the three SERS spectra, resistance genes, and virulence genes and the phenotypic drug resistance information. Finally, the difficult-to-detect phenotypic drug resistance information is converted into SERS spectra, resistance genes, and virulence genes, so as to obtain more accurate drug resistance results with a simpler measurement method.

[0035] (4) The Escherichia coli antimicrobial resistance analysis simulation model provided by the present invention can capture the changes in SERS spectral characteristics of Escherichia coli with different antimicrobial phenotypes and classify them using machine learning algorithms, thereby achieving rapid detection and accurate classification of clinical antimicrobial bacteria.

[0036] (5) The Escherichia coli antimicrobial resistance analysis simulation model provided by the present invention can be used for the detection of real samples and is applicable to multiple fields such as clinical drug sensitivity testing, antimicrobial drug development and hospital infection control. It has important application value and market potential. Attached Figure Description

[0037] Figure 1 The images are TEM images of the synthesized Au@AgNPs, where a is the TEM image of Au@AgNPs, b is the elemental map of Au, c is the elemental map of Ag, and d is the elemental map of Au and Ag.

[0038] Figure 2The images show the Raman spectra of *E. coli* bacteria in real samples, arranged from bottom to top as 1, 2, 3, 4, 5, and 6. 1 represents non-drug-resistant *E. coli*, 2 represents *E. coli* resistant to penicillin, cephalosporins, and quinolones, 3 represents *E. coli* resistant to ciprofloxacin, levofloxacin, ampicillin, and gentamicin, 4 represents *E. coli* resistant to ciprofloxacin, levofloxacin, ampicillin, gentamicin, ceftazidime, cefepime, tetracycline, minocycline, doxycycline, and tobramycin, 5 represents *E. coli* resistant to ciprofloxacin, levofloxacin, ampicillin, cephalosporins, and tetracycline, and 6 represents *E. coli* resistant to ciprofloxacin, levofloxacin, ampicillin, ceftazidime, tetracycline, minocycline, and doxycycline.

[0039] Figure 3 The results of machine learning in detecting drug resistance in E. coli are shown in the figure. A is the machine learning flowchart integrating genotype and spectral data, B is the results of accuracy, precision, recall and F1 score, and C is the area under the curve of the diagnosis of the simulation model for analyzing drug resistance in E. coli. Detailed Implementation

[0040] The present invention will be further described below with reference to specific embodiments. Those skilled in the art will be able to implement the present invention based on these descriptions. Furthermore, the embodiments of the present invention described below are generally only some, not all, of the embodiments of the present invention. Therefore, all other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0041] Example 1: Simulation Model Establishment

[0042] (1) Preparation of SERS substrate material

[0043] Synthesis of gold nanoparticles: 100 mL of ultrapure water and 1 mL of 1% (w / v) chloroauric acid were mixed and heated to boiling while stirring constantly. After boiling, 1.0 mL of 1% (w / v) sodium citrate solution was quickly added. The solution changed from yellow to deep red, indicating successful formation of gold nanoparticles. After stirring for another 15 minutes, the mixture was allowed to cool naturally to room temperature to obtain a gold nanoparticle solution.

[0044] Silver-coated gold nanoparticles: 3 mL of gold nanoparticle solution was placed in a shaker (25℃, 300 rpm), and 125 μL of 10 mmol / L silver nitrate and 125 μL of 10 mmol / L ascorbic acid were slowly added sequentially for silver ion coating and reduction. The color changed from purplish-red to orange-yellow, indicating the formation of a core-shell structure. The reaction was continued with shaking for 20 minutes to ensure the formation of complete silver-coated gold nanoparticles, denoted as Au@Ag NPs.

[0045] TEM tests were performed on Au@Ag NPs, and the results are as follows: Figure 1 As shown, silver is uniformly coated on the surface of the synthesized gold nanoparticles, proving that a silver-coated gold nanomaterial has been synthesized.

[0046] (2) Simulation model for analyzing drug resistance in Escherichia coli

[0047] S1. Obtaining Escherichia coli bacterial suspension:

[0048] In this embodiment, the Escherichia coli bacteria were collected from real samples (such as patients' blood, urine, and sputum). The phenotypic drug resistance of the real samples collected in this embodiment showed resistance to at least one of the following 22 antibiotics. The 22 antibiotics are: Amikacin (AMK), Tobramycin (TOB), Gentamicin (GEN), Ampicillin (AMP), Piperacillin (PIP), Polymyxin E (COL), Doxycycline (DOX), Tetracycline (TET), Minocycline (MIN), Ertapenem (ERT), Nitrofurantoin (NIT), Cefoperazone (CFP / SUL), Ampicillin-Sulbactam (SAM), Piperacillin (TZP), Sulfamethoxazole (SXT), Ciprofloxacin (CIP), Levofloxacin (LVX), Carbapenem-resistant Gram-negative bacilli (CRO), Cefepime (FEP), Cefotaxime (CTX), Ceftazidime (CAZ), and Aztreonam (ATM).

[0049] The methods for collecting real samples are as follows:

[0050] Methods for collecting pure colonies: After collecting blood, urine, and sputum samples, bacterial cultures were performed according to routine microbiological methods. Blood samples were inoculated into blood culture bottles and incubated in an automated culture system at 37°C; urine samples were inoculated onto blood agar or MacConkey medium using a quantitative inoculation loop and incubated at 37°C for 18–24 hours; sputum samples, after being screened under a microscope, were inoculated into suitable selective media and incubated in a constant temperature incubator at 37°C, and typical colonies were observed for subsequent identification and experimental use.

[0051] SERS spectral acquisition method for bacteria (based on 20 acquisitions, total acquisition time 40 min): To enrich pathogens in complex samples, differential centrifugation combined with cell disruption is used. Samples (such as blood or sputum) are diluted with PBS and then briefly disrupted using erythrocyte lysis buffer or sonication to release latent bacteria. Subsequently, differential centrifugation is performed sequentially: first at low speed (300×g, 5 min) to remove large cell debris, then at medium speed (3000×g, 10 min) to further remove host cell residues, and finally at high speed (10,000×g, 10–15 min) to precipitate bacterial particles. The precipitate is washed with PBS and resuspended for SERS detection or other analytical procedures.

[0052] For urine sample processing: First, collect 5 mL of midstream clean urine and filter it using a 5 μm filter membrane to remove detached cells and impurities. After filtration, the sample is centrifuged (8000 × g, 10 min) to enrich bacteria. The supernatant is discarded, and the precipitate is washed 1–2 times with PBS to remove background interference. Finally, the bacteria are resuspended in a small volume of PBS for subsequent SERS detection or other analytical procedures. This method can efficiently obtain pathogenic bacteria in urine under aseptic culture conditions.

[0053] For sputum sample processing: Morning deep cough sputum is preferred. After removing saliva contamination, dilute with an equal volume of PBS and homogenize thoroughly. After coarse filtration (30µm filter membrane) to remove large impurities, centrifuge (5000×g, 10min) to enrich pathogens. Wash the precipitate 1–2 times with PBS. If necessary, short-duration sonication or enzymatic methods (such as DNase or proteinase K) can be used to assist in removing mucus components and improve bacterial purity. Finally, resuspend the bacteria for SERS detection or subsequent experiments.

[0054] After collecting the real samples, the bacteria in the real samples were cultured as follows: Real samples frozen at -80℃ were aseptically inoculated onto blood agar plates and incubated overnight at 37℃. The real samples cultured overnight were used for purity analysis. Single-clone strains were selected and cultured in LB liquid medium for 18-22 h to ensure that the *E. coli* bacteria in the real samples were in the logarithmic growth phase. 1 mL of the treated real sample was centrifuged at 10000 rpm for 10 min and the supernatant was discarded. The sample was resuspended in 5 mL of deionized water and centrifuged at 10000 rpm for 5 min, and the supernatant was discarded. This step was repeated twice. The precipitate was then resuspended in 100 μL of deionized water to obtain the desired *E. coli* bacterial suspension.

[0055] S2. Mix 20 μL of Au@Ag NPs solution with equal volumes of Escherichia coli bacterial suspensions cultured in S1, and incubate at room temperature for 10 min to ensure thorough mixing of Escherichia coli bacteria with the silver-coated gold nanoparticle SERS substrate to obtain a mixed solution.

[0056] S3. SERS Spectral Data: 5 μL of the mixture obtained in S2 was dropped onto a silicon wafer, dried at room temperature, and then SERS spectra were acquired using a Raman spectrometer. The Raman spectrometer parameters were: range 600–1800 cm⁻¹. -1 The integration time is 2 seconds, the laser wavelength is 785 nm, and the resolution is 3.5 cm. -1 The laser power was 350 mW, the laser spot size was 100 μm, and the cumulative duration was 5 times.

[0057] Each E. coli sample underwent at least 20 tests to obtain 20 spectra, ensuring spectral repeatability and accuracy. SERS spectra are as follows: Figure 2 As shown, using silver-coated gold nanoparticles as the SERS substrate material amplifies the Raman signals of specific molecules on the surface of *E. coli* bacteria, such as surface proteins, phospholipids, secretions, and some nucleic acids. Subsequently, each SERS spectrum undergoes baseline correction and normalization to obtain the final SERS spectral data for each *E. coli* bacterial sample.

[0058] S4. Resistance and Virulence Gene Data: The results of resistance genes and virulence genes in drug-resistant *E. coli* were obtained using PCR sequencing technology (PCR sequencing time 110 min). The *E. coli* bacterial suspensions from S1 were sequenced to obtain the presence of resistance and virulence genes in each sample. Gene expression was recorded as 1, and no expression as 0. The resistance and virulence gene results were binarized to obtain resistance and virulence gene data.

[0059] S5. Perform drug susceptibility testing on the real samples involved in S1 to obtain the drug resistance results of Escherichia coli corresponding to each sample, and use them as labeled Escherichia coli drug resistance data.

[0060] S6. For example Figure 3 The process shown in A involves inputting SERS spectral data into a CNN model and resistance gene data and virulence gene data into an MLP model; using tagged E. coli drug resistance data to train and validate the CNN and MLP models, and then fusing the CNN and MLP models to obtain a simulation model for analyzing E. coli bacterial drug resistance.

[0061] During the fusion process, resistance gene data and virulence gene data are fused with SERS spectral data respectively. During fusion, the data in resistance gene data and virulence gene data that have poor correlation with SERS spectral data are removed by Lasso regression cleaning, and the remaining data continue to be fused with SERS spectral data.

[0062] The results of the gene data cleaning are as follows:

[0063] Resistance genes before selection: CTX-M, TEM, OXA, DHA, aph(3'')-Ib, aph(6)-Id, aadA5, aph(3')-IIa, aac(3)-II, mph(A), erm(B), qnr, qepA, sul2, sul1, sul3, tet(A), fosA3, floR, cmlA1, dfrA17, qacE.

[0064] The selected resistance genes are: CTX-M, TEM, OXA, aadA5, aac(3)-II, mph(A), erm(B), sul2, tet(A), fosA3, floR and dfrA17).

[0065] Virulence genes before selection: fim, cag / cgs, ecp, kps, dra, cfa, pap, ompA, fdec,sat, tsh, fha, ibe, neu, GSP, ent, fep, iuc, iutA, fur, chu, iron, ybt, hcp / tssD, verG / tssl, clb, senB, tss, irp1 / irp2 / fyuA, espL, fes.

[0066] The selected virulence genes are: cfa, fha, neu, iron, clb, senB, and tss.

[0067] Example 2 Model Validation

[0068] The method provided in Example 1 and the obtained simulation model for analyzing the drug resistance of Escherichia coli were used to test the urine sample. The test analysis time was 2 minutes, and the phenotypic drug resistance results of Escherichia coli in the urine sample were obtained.

[0069] At the same time, drug susceptibility testing was used to determine the phenotypic drug resistance results of the same "urine sample" of E. coli.

[0070] The results obtained from the simulation model of Escherichia coli antibiotic resistance analysis were compared with the results obtained from the drug susceptibility test to obtain the prediction accuracy of antibiotic resistance phenotypes.

[0071] Table 1: Prediction accuracy of different antibiotic resistance phenotypes based on CNN and MLP models

[0072]

[0073] In addition, the accuracy, precision, and recall of the simulation model for analyzing Escherichia coli antibiotic resistance were tested and evaluated, and the results are as follows: Figure 3 As shown in B, the accuracy and precision of the simulation model for analyzing drug resistance in Escherichia coli are close to 100%.

[0074] Table 2: Comparison of Detection Time

[0075]

[0076] Based on the calculations regarding the time required to obtain data in the model establishment of Example 1 and the model verification of Example 2, it can be found that the method provided by this application can complete data processing and result acquisition in about 2.5 hours; while conventional drug sensitivity tests require 24 hours. Comparing these results, this application is clearly more advantageous, enabling rapid diagnosis and accurate results.

[0077] Example 3: Testing and Use

[0078] A urine sample was selected as the test subject, and the test was conducted using the protocol provided in Example 1 and the obtained simulation model for analyzing the antibiotic resistance of E. coli. The results are shown in Table 3. When the probability value of the predicted resistance is greater than 0.5, the E. coli is considered to be resistant to this antibiotic.

[0079] Table 3: Results of Antimicrobial Resistance Prediction

[0080]

[0081] Comparative Example 1: SERS Substrate Material Screening

[0082] This comparative example is basically the same as Example 1, except that the SERS substrate material used is replaced with gold nanoparticles.

[0083] Synthesis of gold nanoparticles: 100 mL of ultrapure water was mixed with 0.25 mL of 1% (w / v) chloroauric acid and heated to boiling while stirring constantly. After boiling, 1.0 mL of 1% (w / v) sodium citrate solution was quickly added. The solution changed from yellow to deep red, indicating successful formation of gold nanoparticles. After stirring for another 15 minutes, the mixture was allowed to cool naturally to room temperature to obtain a gold nanoparticle solution.

[0084] Table 4: Enhancement effects of different nanomaterials as SERS substrates

[0085]

[0086] As can be seen from Table 4, when the substrate material is replaced with gold (Au) nanoparticles, the detection accuracy decreases significantly.

[0087] Comparative Example 2: Model and Input Data Selection

[0088] During the model training phase, the input data and the model used significantly impact the final training results. Specifically, a CNN model was used to train the SERS spectral data alone, while an MLP model was used to train the resistance gene and virulence gene data. When inputting SERS spectral data, resistance gene data, and virulence gene data, a dual-model approach using both CNN and MLP was employed. The CNN model was used to process the SERS spectral data, while the MLP model was used to process the resistance gene and virulence gene data.

[0089] Apart from the variables mentioned above, the other contents are the same as in Example 1. The accuracy rates corresponding to different models and input data after changes are shown in Table 5.

[0090] Observations show that when a CNN model is used to process SERS spectral data and an MLP model is used to process resistance gene data and virulence gene data respectively for model training, the final prediction accuracy is higher than that of using a single CNN model. Furthermore, using three types of data—SERS spectral data, resistance gene data, and virulence gene data—results in a significantly higher prediction accuracy than using a single phenotypic data or a single gene data.

[0091] Table 5: Phenotypic Antibiotic Prediction Accuracy of Different Models and Input Data

[0092]

Claims

1. A method for detecting antibiotic resistance of E. coli based on SERS and machine learning, characterized in that, The application relates to a method for analyzing drug resistance of escherichia coli, comprising the following steps: S1.1 obtaining SERS spectra of drug-resistant escherichia coli, and normalizing the SERS spectra to obtain SERS spectrum data; obtaining resistance gene results and virulence gene results of the drug-resistant escherichia coli, and respectively binarizing gene expression to obtain resistance gene data and virulence gene data; S1.2 inputting the SERS spectrum data of S1.1 into a CNN model, inputting the resistance gene data of S1.1 and the virulence gene data of S1.1 into an MLP model, training and verifying the CNN model and the MLP model by using labeled drug resistance data of escherichia coli, and fusing the CNN model and the MLP model to obtain an escherichia coli drug resistance analysis simulation model; S1.3 obtaining SERS spectra of the drug-resistant escherichia coli to be detected, and normalizing the SERS spectra to obtain SERS spectrum data; obtaining resistance gene results and virulence gene results of the drug-resistant escherichia coli to be detected, and respectively binarizing gene expression to obtain resistance gene data and virulence gene data; S1.4 inputting the SERS spectrum data, the resistance gene data and the virulence gene data obtained in S1.3 into the escherichia coli drug resistance analysis simulation model to obtain drug resistance results of the target drug-resistant escherichia coli.

2. The detection method of claim 1, wherein, In S1.1 and S1.3, the SERS spectra are baseline corrected and then normalized, and each SERS spectrum is measured at least 20 times.

3. The method of claim 1, wherein the detecting is performed by a method selected from the group consisting of mass spectrometry, nuclear magnetic resonance, and chromatography. In S1.1 and S1.3, the SERS spectra are measured by mixing silver-coated gold nanoparticles with the drug-resistant escherichia coli and using SERS technology.

4. The detection method of claim 3, wherein, The test parameters of the SERS spectrum are: range 600~1800 cm -1 , integration time 2~3 seconds, laser wavelength 785 nm, resolution 3~4 cm -1 , laser power 350~380 mW, laser spot 100~120 μm, cumulative time 5 times.

5. The detection method as described in claim 3, characterized in that, The preparation method of the silver-coated gold nanoparticles is as follows: S5.1 under heating conditions, mixing a chloroauric acid solution and a sodium citrate solution, and after the solution changes from yellow to deep red, stirring for 10-20 min, and cooling to obtain a gold nanoparticle suspension; S5.2 adding ascorbic acid and silver nitrate into the gold nanoparticle suspension in sequence, oscillating for 20-30 min after the suspension color changes from purple red to orange yellow, and obtaining a silver-coated gold nanoparticle solution.

6. The detection method as described in claim 1, characterized in that, In S1.2, the labeled drug resistance data of escherichia coli is the escherichia coli phenotype drug resistance data obtained by a drug sensitivity experiment.

7. The detection method as described in claim 1, characterized in that, In S1.1, the drug-resistant escherichia coli is separated from a biological sample; and in S1.3, the drug-resistant escherichia coli to be detected is separated from a biological sample.

8. The detection method of claim 7, wherein, The biological sample is selected from blood, urine, sputum or wound secretion.

9. The detection method as described in claim 1, characterized in that, In S1.1 and S1.3, the resistance gene results of the drug-resistant escherichia coli and the virulence gene results of the drug-resistant escherichia coli are detected by a sequencing technology.

10. The method of claim 1, wherein, In S1.2, in the fusion process, the resistance gene data and the virulence gene data are respectively fused with the SERS spectrum data, and the parts of the resistance gene data and the virulence gene data with poor correlation with the SERS spectrum data are removed during the fusion, and the remaining data continue to be fused with the SERS spectrum data.

Images