A method and system for predicting the antimicrobial resistance phenotype of Escherichia coli from clinical pets

Abstract

The present invention discloses a method and system for predicting the antibiotic resistance phenotype of clinical pet-derived Escherichia coli, which belongs to the field of intelligent medical testing. The MIC test of antibiotics and whole genome sequencing are performed on pet-derived Escherichia coli samples to obtain MIC classification labels and pan-genome features, respectively; the pan-genome features related to the MIC classification labels are screened to obtain the optimal feature combination; a feedforward neural network model is constructed and optimized to obtain an optimized prediction model for antibiotics; the optimal feature combination is input into the optimized prediction model for antibiotics, and the MIC prediction results of antibiotics are output; the MIC prediction results are converted into SIR labels to determine the resistance phenotype of antibiotics. The present invention screens out the pan-genome related to the resistance phenotype from multiple angles as a feature input model, and simultaneously predicts the MIC label and the SIR labels of the three categories of sensitive, intermediate and resistant, to provide assistance for the diagnosis and treatment of bacterial infections in pet clinics and subsequent research.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical testing technology, and more specifically to a method and system for predicting the antibiotic resistance phenotype of clinical pet-derived Escherichia coli. Background Art

[0002] In the field of pet medicine, the widespread use of antimicrobial drugs as an important means of treating infectious diseases has become one of the important driving forces for the development of bacterial resistance. In other words, the frequent and possibly inappropriate use of antimicrobial drugs in pet clinics is a core factor in aggravating the phenomenon of bacterial resistance.

[0003] At present, the resistance rate of E. coli infected by dogs and cats to multiple antibiotics is higher than that of humans. The existing drug sensitivity test requires a long period of time and high technical requirements. The developed test kits often only target specific resistance genes and are not enough for promotion. The developed bacterial resistance phenotype prediction model has not yet been focused on the construction of pet clinical samples, and its generalization is insufficient. At the same time, the existing model lacks feature engineering, so many meaningless features are included in the model, which is not conducive to subsequent clinical and scientific research.

[0004] Therefore, it is an urgent problem for technical personnel in this field to propose a method and system for predicting the antimicrobial resistance phenotype of Escherichia coli from clinical pets and improve the accuracy of resistance phenotype prediction. Summary of the invention

[0005] In view of this, the present invention provides a method and system for predicting the antibiotic resistance phenotype of Escherichia coli from clinical pets.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] In one aspect, the present invention discloses a method for predicting the antimicrobial resistance phenotype of Escherichia coli from clinical pets, comprising the following steps:

[0008] Obtaining a pet-derived Escherichia coli sample, performing an antimicrobial MIC test and whole genome sequencing on the pet-derived Escherichia coli sample, and obtaining a MIC classification label and a pan-genome feature, respectively;

[0009] Screening pan-genomic features related to the MIC classification label to obtain an optimal feature combination;

[0010] Construct and optimize the feedforward neural network model to obtain the optimization prediction model of antimicrobial drugs;

[0011] Inputting the optimal feature combination into the antimicrobial drug optimization prediction model, and outputting the MIC prediction result of the antimicrobial drug;

[0012] The MIC prediction result is converted into a SIR signature, and the drug resistance phenotype of the antimicrobial drug is determined based on the SIR signature.

[0013] Preferably, screening pan-genomic features related to the MIC classification label to obtain an optimal feature combination includes:

[0014] Preliminary screening of the pan-genome features was performed based on variance analysis to obtain preliminary screening features;

[0015] Performing Fisher's exact test on the preliminary screening features, and retaining the second screening features that meet the preset conditions;

[0016] The second screening features are ranked according to the mutual information analysis scores with the MIC categories and the top 10% are selected as candidate features;

[0017] Based on the candidate features, a genetic algorithm is used to determine the optimal feature combination for the antimicrobial phenotype prediction task.

[0018] Preferably, the pan-genome features are preliminarily screened based on variance analysis to obtain preliminary screening features, including:

[0019] Based on variance analysis, pan-genome features that exceeded the preset value or were carried by less than a number of samples were removed, and the remaining pan-genome features were used as preliminary screening features.

[0020] Preferably, the genetic algorithm is implemented based on random forest.

[0021] Preferably, constructing and optimizing a feedforward neural network model to obtain an antimicrobial drug optimization prediction model includes:

[0022] Determine the hyperparameters of the feedforward neural network model through Bayesian optimization to obtain an optimized model;

[0023] Based on Monte Carlo cross-validation, several groups of training and testing data sets were randomly generated, and the optimization model was trained to obtain the optimized prediction model of antimicrobial drugs.

[0024] On the other hand, the present invention also discloses a clinical pet-derived Escherichia coli antibiotic resistance phenotype prediction system, comprising:

[0025] A sample processing module is used to obtain a pet-derived Escherichia coli sample, perform MIC testing of antimicrobial drugs and whole genome sequencing on the pet-derived Escherichia coli sample, and obtain a MIC classification label and a pan-genome feature, respectively;

[0026] A feature screening module, used to screen pan-genomic features related to the MIC classification label to obtain an optimal feature combination;

[0027] Model optimization module, used to construct and optimize the feedforward neural network model to obtain the optimized prediction model of antimicrobial drugs;

[0028] A MIC prediction module, used for inputting the optimal feature combination into the antimicrobial drug optimization prediction model, and outputting the MIC prediction result of the antimicrobial drug;

[0029] The drug resistance phenotype determination module is used to convert the MIC prediction result into a SIR label and determine the drug resistance phenotype of the antimicrobial drug based on the SIR label.

[0030] Preferably, the feature screening module includes:

[0031] A preliminary screening unit, used for performing preliminary screening on the pan-genome features based on variance analysis to obtain preliminary screening features;

[0032] A second screening unit, used to perform Fisher's exact test on the preliminary screening features and retain the second screening features that meet the preset conditions;

[0033] A candidate feature screening unit, used to sort the second screening features according to the mutual information analysis scores with the MIC category and select the top 10% as candidate features;

[0034] The optimal feature combination confirmation unit is used to determine the optimal feature combination for the antimicrobial drug phenotype prediction task based on the candidate features using a genetic algorithm.

[0035] Through the above technical solutions, it can be known that compared with the prior art, the present invention discloses a method and system for predicting the antibiotic resistance phenotype of Escherichia coli from clinical pets, performs MIC testing and whole genome sequencing of antibiotics on pet-derived Escherichia coli samples, and obtains MIC classification labels and pan-genome features respectively; screens pan-genome features related to MIC classification labels to obtain the optimal feature combination; constructs and optimizes a feedforward neural network model to obtain an optimized prediction model for antibiotics; inputs the optimal feature combination into the optimized prediction model for antibiotics, and outputs the MIC prediction results of antibiotics; converts the MIC prediction results into SIR labels, and determines the resistance phenotype of antibiotics based on the SIR labels. The present invention selects the pan-genome related to the resistance phenotype as a feature input model from multiple angles, and simultaneously predicts the MIC label and the SIR labels of the three categories of sensitive, intermediate and resistant, to provide assistance for the diagnosis and treatment of bacterial infections in pet clinics and subsequent research. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0037] Figure 1 A flow chart of the method provided by the present invention;

[0038] Figure 2 A schematic diagram of the accuracy of training sets and test sets of different antibacterial drugs based on Monte Carlo cross-validation provided in an embodiment of the present invention;

[0039] Figure 3 Schematic diagram of micro-average AUC of different antimicrobial drug prediction models;

[0040] Figure 4(a), Figure 4(b), Figure 4(c), and Figure 4(d) are schematic diagrams showing the comparison of the accuracy, macro F1 score, macro precision, and macro recall of different antimicrobial prediction models after converting the MIC prediction results into SIR labels;

[0041] Figure 5 This is a schematic diagram of the prediction accuracy of different antimicrobial drugs for drug-resistant strains;

[0042] Figure 6(a), Figure 6(b), and Figure 6(c) are the pan-genomic feature screening results of meropenem, ampicillin, and amikacin, respectively;

[0043] Figure 7 It is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] refer to Figure 1 The embodiment of the present invention discloses a method for predicting the antimicrobial resistance phenotype of Escherichia coli from clinical pets, comprising the following steps:

[0046] S1. Obtain pet-derived Escherichia coli samples, perform antimicrobial MIC testing and whole genome sequencing on the pet-derived Escherichia coli samples, and obtain MIC classification labels and pan-genome features, respectively.

[0047] In this example, 1010 clinical canine and feline Escherichia coli samples were collected to construct a data set. These samples cover Escherichia coli isolated from dogs and cats under different clinical infection conditions. In order to ensure the consistency and accuracy of the data, a unified minimum inhibitory concentration (MIC) test was performed, and the resulting MIC value was used as a label for each sample. In addition, the whole genome sequence (WGS) of each isolate was annotated, and the pan-genome was used as a feature.

[0048] The isolated E. coli were revived on brain-heart perfusion agar at 37°C for 18–24 h and reconfirmed by matrix-assisted laser desorption ionization time-of-flight mass spectrometry and 16S rRNA gene sequencing. The antimicrobial susceptibility of E. coli isolates was tested by broth microdilution method using specially prepared broth microdilution plates (Thermo Fisher Scientific) according to the Clinical and Laboratory Standards Institute (CLSI) document M07. The results were further interpreted according to the breakpoints in CLSI VET01S-Ed5 and CLSIM100-Ed31.

[0049] In this example, 9 types of antimicrobial drugs were selected, including penicillins (ampicillin (AMP)), penicillins and β-lactamase inhibitors (amoxicillin-clavulanate (AMC)), third-generation and fourth-generation cephalosporins (cefotaxime (CRO) and cefquinome (CEQ)), carbapenems (meropenem (MEM)), aminoglycosides (amikacin (AMK) and gentamicin (GEN)), fluoroquinolones (levofloxacin (LVX) and enrofloxacin (ENR)), sulfonamides (trimethoprim-sulfamethoxazole (SXT)), tetracyclines (doxycycline (DOX)) and chloramphenicol (florfenicol (FFC)), and Escherichia coli ATCC25922 was used as the quality control strain.

[0050] S2. Screen pan-genomic features related to MIC classification labels to obtain the optimal feature combination, including:

[0051] S21. Perform preliminary screening of pan-genome features based on variance analysis to obtain preliminary screening features.

[0052] Specifically, based on variance analysis, pan-genome features that exceed a preset value or are carried by less than a number of samples are removed, and the remaining pan-genome features are used as preliminary screening features. In this embodiment, pan-genome features that are carried by more than 85% or less than 10 samples are removed.

[0053] S22. Perform Fisher's exact test on the preliminary screening features and retain the second screening features that meet the preset conditions.

[0054] From a probability perspective, the preliminary screening features were first processed using the Fisher exact test, and the features that were significantly correlated with each antimicrobial drug MIC signature (P <= 0.05) were retained as the second screening features.

[0055] S23. Sort the second screening features according to the mutual information analysis scores with the MIC categories and select the top 10% as candidate features.

[0056] S24. Based on the candidate features, a genetic algorithm is used to determine the optimal feature combination for the antimicrobial drug phenotype prediction task, wherein the genetic algorithm is implemented based on random forest.

[0057] In this embodiment, the genetic algorithm is implemented in Python. The population size is initially set to 50, and each individual represents a different combination of pan-genomic features. According to the RF evaluation criteria, the top 30% of the best performing individuals are selected as the next generation. In the next generation, 2 individuals are randomly selected as parents for hybridization with a hybridization rate of 0.6 to reproduce offspring until the population size is restored. Then, all offspring are mutated with a mutation rate of 0.3. This process was repeated for 50 iterations, retaining the best feature combination of each iteration. Random forests were constructed using "RandomForestClassifier" in sklearn (version 1.1.1), with "n_estimators" set to 30000. A 5-fold cross validation was performed using "cross_val_score" to evaluate the results.

[0058] The feature screening method based on S2 can more reasonably screen out features related to drug resistance phenotypes. For example, aac(6')-Ib and rmtB1 of amikacin (Figure 6(c)), bla of meropenem NDM-5 (Figure 6(a)), bla of ampicillin TEM-1 (Figure 6(b)), which can help in-depth clinical analysis or research to discover new drug-resistant genes / mechanisms.

[0059] S3. Construct and optimize the feedforward neural network model to obtain the optimized prediction model of antimicrobial drugs.

[0060] S4. Input the optimal feature combination into the antimicrobial drug optimization prediction model and output the MIC prediction result of the antimicrobial drug.

[0061] S5. Convert the MIC prediction results into SIR signatures and determine the resistance phenotype of antimicrobial drugs based on the SIR signatures.

[0062] This example uses a feedforward neural network as the core deep learning model and takes the optimal feature combination as input. Hyperparameters such as the number of neurons, number of hidden layers, and learning rate are determined through Bayesian optimization, and then the best training set-test set combination is selected based on 100 Monte Carlo cross-validation to further evaluate the accuracy of the model ( Figure 2 ) and micro-average AUC( Figure 3 ),according to Figure 3 It can be seen that all models have good overall classification capabilities.

[0063] Considering the actual clinical application, this example also converts the actual and model-predicted MIC labels into SIR (susceptible (S), intermediate (I) and resistant (R)) labels according to the CLSI standard and evaluates the accuracy, macro F1 score, macro precision and macro recall of the model for SIR labels, as shown in Figure 2. Figure 4(a)-Figure 4(d) As shown, some drugs such as doxycycline (DOX) have been greatly improved, and the prediction accuracy of 10 drugs exceeds 70%.

[0064] This example also evaluates the prediction accuracy of clinical drug-resistant strains, referring to Figure 5 ,For each drug-resistant strain classified into the test set, the prediction accuracy of the 10 drugs for the drug-resistant strains ranged from 78.95% to 99.07%.

[0065] During the implementation, the feedforward neural network model was developed using TensorFlow (version 2.11.1). In order to optimize the hyperparameters of the model, the hyperparameter optimization tool hyperopt (version 0.2.7) was used. Hyperopt uses a Bayesian optimization algorithm called the tree-structured Parzen Estimator, which effectively searches the parameter space by estimating and optimizing the probability distribution of parameters to find the optimal hyperparameter combination. The judgment criteria used a ten-fold cross-validation.

[0066] The following hyperparameter space is defined: the number of hidden layers ranges from 3 to 8, the number of neurons per layer ranges from 100 to 400, the learning rate ranges from 0.0001 to 0.01, and the batch size ranges from 20 to 100. Bayesian optimization is used and the maximum number of evaluations is set to 100 to ensure that the best hyperparameter combination is found within a limited time frame.

[0067] Finally, based on Monte Carlo cross-validation, 100 sets of training and test data sets were randomly generated, with a training-test ratio of 7:3 for each set. The feedforward neural network model was trained using the optimized hyperparameters, and the Adam optimizer was used for learning rate optimization, with "categorical_crossentropy" as the loss function and accuracy as the evaluation indicator.

[0068] On the other hand, Figure 7 As shown, the present invention also discloses a clinical pet-derived Escherichia coli antibiotic resistance phenotype prediction system, comprising:

[0069] A sample processing module is used to obtain pet-derived E. coli samples, perform MIC tests on antimicrobial drugs and perform whole-genome sequencing on the pet-derived E. coli samples to obtain MIC classification labels and pan-genome features, respectively;

[0070] Feature screening module, used to screen pan-genomic features related to MIC classification labels to obtain the optimal feature combination;

[0071] Model optimization module, used to construct and optimize the feedforward neural network model to obtain the optimized prediction model of antimicrobial drugs;

[0072] MIC prediction module, used to input the optimal feature combination into the antimicrobial drug optimization prediction model and output the MIC prediction results of the antimicrobial drugs;

[0073] The drug resistance phenotype determination module is used to convert the MIC prediction results into SIR labels and determine the drug resistance phenotype of the antimicrobial drug based on the SIR labels.

[0074] Preferably, the feature screening module includes:

[0075] A preliminary screening unit, used to perform preliminary screening of pan-genome features based on variance analysis to obtain preliminary screening features;

[0076] A second screening unit is used to perform Fisher's exact test on the preliminary screening features and retain the second screening features whose correlation with the MIC tag meets the preset conditions;

[0077] A candidate feature screening unit, used to sort the second screening features according to the mutual information analysis scores with the MIC categories and select the top 10% as candidate features;

[0078] The optimal feature combination confirmation unit is used to determine the optimal feature combination for the antimicrobial drug phenotype prediction task based on candidate features using a genetic algorithm.

[0079] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0080] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting the antimicrobial resistance phenotype of Escherichia coli from clinical pets, characterized in that: The following steps are involved: Obtaining a pet-derived Escherichia coli sample, performing an antimicrobial MIC test and whole genome sequencing on the pet-derived Escherichia coli sample, and obtaining a MIC classification label and a pan-genome feature, respectively; Screening pan-genomic features related to the MIC classification label to obtain an optimal feature combination; Construct and optimize the feedforward neural network model to obtain the optimization prediction model of antimicrobial drugs; Inputting the optimal feature combination into the antimicrobial drug optimization prediction model, and outputting the MIC prediction result of the antimicrobial drug; Converting the MIC prediction result into a SIR signature, and determining the drug resistance phenotype of the antimicrobial drug based on the SIR signature; Screening pan-genomic features related to the MIC classification label to obtain the optimal feature combination, including: Preliminary screening of the pan-genome features was performed based on variance analysis to obtain preliminary screening features; Performing Fisher's exact test on the preliminary screening features, and retaining the second screening features that meet the preset conditions; The second screening features are ranked according to the mutual information analysis scores with the MIC categories and the top 10% are selected as candidate features; Based on the candidate features, a genetic algorithm is used to determine the optimal feature combination for the antimicrobial drug phenotype prediction task, and the genetic algorithm is implemented based on random forest.

2. A method for predicting the antimicrobial resistance phenotype of Escherichia coli from clinical pets according to claim 1, characterized in that: The pan-genome features were preliminarily screened based on variance analysis to obtain preliminary screening features, including: Based on variance analysis, pan-genome features that exceeded the preset value or were carried by less than a number of samples were removed, and the remaining pan-genome features were used as preliminary screening features.

3. The method for predicting the antimicrobial resistance phenotype of Escherichia coli from clinical pets according to claim 1, characterized in that: Construct and optimize the feedforward neural network model to obtain the optimization prediction model of antimicrobial drugs, including: Determine the hyperparameters of the feedforward neural network model through Bayesian optimization to obtain an optimized model; Based on Monte Carlo cross-validation, several groups of training and testing data sets were randomly generated, and the optimization model was trained to obtain the optimized prediction model of antimicrobial drugs.

4. A clinical pet-derived Escherichia coli antibiotic resistance phenotype prediction system, characterized in that: include: A sample processing module is used to obtain a pet-derived Escherichia coli sample, perform MIC testing of antimicrobial drugs and whole genome sequencing on the pet-derived Escherichia coli sample, and obtain a MIC classification label and a pan-genome feature, respectively; A feature screening module, used to screen pan-genomic features related to the MIC classification label to obtain an optimal feature combination; Model optimization module, used to construct and optimize the feedforward neural network model to obtain the optimized prediction model of antimicrobial drugs; A MIC prediction module, used for inputting the optimal feature combination into the antimicrobial drug optimization prediction model, and outputting the MIC prediction result of the antimicrobial drug; A drug resistance phenotype determination module, used to convert the MIC prediction result into a SIR tag, and determine the drug resistance phenotype of the antimicrobial drug based on the SIR tag; The feature screening module includes: A preliminary screening unit, used for performing preliminary screening on the pan-genome features based on variance analysis to obtain preliminary screening features; A second screening unit, used to perform Fisher's exact test on the preliminary screening features and retain the second screening features that meet the preset conditions; A candidate feature screening unit, used to sort the second screening features according to the mutual information analysis scores with the MIC category and select the top 10% as candidate features; The optimal feature combination confirmation unit is used to determine the optimal feature combination for the antimicrobial drug phenotype prediction task based on the candidate features using a genetic algorithm, wherein the genetic algorithm is implemented based on a random forest.

Images