Microbial marker for cervical cancer and application thereof

Abstract

The invention provides a plurality of microbial markers for cervical cancer and an application of the microbial markers. The microbial markers comprise 11 bacteria, namely, Porphyromonas aeruginosa, Lactobacillus crispatus, Klebsiella pneumoniae, Gemerella haemolysms and the like, and the microbial markers comprise 11 bacteria, namely, the Porphyromonas aeruginosa, the Lactobacillus crispatus, the Klebsiella pneumoniae, the Gemerella haemolysms and the like. According to the method, the threshold value of the output risk value is obtained according to the trained random forest model, and then the probability value that the cervical lesion sample is the high-risk lesion can be obtained only by obtaining the relative abundance of the screened 11 genital tract microbial markers and inputting the relative abundance into the established random forest model. The machine learning model is constructed based on 16S sequencing, the high risk of cervical cancer can be recognized in time, the method is used for auxiliary diagnosis of cervical cancer patients, early discovery, early prevention and early treatment of cervical cancer are facilitated, and the accessibility of cervical cancer screening is improved.

Description

Technical Field

[0001] The present invention belongs to the field of biotechnology and microorganisms, and specifically relates to a variety of cervical cancer microbial markers and applications thereof. Background Art

[0002] Cervical cancer is a common malignant tumor that seriously threatens women's health and has become a major global public health issue. According to data from the World Health Organization (WHO) in 2020, there are approximately 604,000 new cases of cervical cancer and 342,000 deaths worldwide, of which 88.1% of new cases and 91.4% of deaths occur in middle- and low-income countries. In recent years, the incidence of cervical cancer in my country has been increasing year by year. In 2021, there were 111,820 new cases of cervical cancer in my country, accounting for 18% of new cases worldwide, and 61,579 deaths, making China the second largest country in the world in terms of cervical cancer burden. The average age of onset is gradually decreasing, with a trend of younger age. Cervical cancer often has an insidious onset, no symptoms in the early stages, and is often discovered in the late stages. There are fewer treatment options and a poor prognosis, while the 5-year survival rate of patients with early cervical cancer can reach 92%. As the only malignant tumor with a clear cause that can be eliminated through tertiary prevention measures, early detection of cervical cancer is of great significance for the intervention and treatment of cervical cancer.

[0003] In recent years, many studies have pointed out that vaginal microorganisms play an important role in the occurrence and development of HPV infection, cervical intraepithelial neoplasia and cervical cancer. When the vaginal microecology is unbalanced and the number and activity of lactobacilli are reduced, HPV is more likely to act on cervical epithelial cells, leading to cervical intraepithelial neoplasia and increasing the risk of cervical cancer.

[0004] At present, the conventional screening methods for cervical lesions in my country are HPV testing and cervical cytology, but both methods have certain limitations. Cytology has high specificity and positive predictive value, but low sensitivity, which can easily lead to missed diagnosis; HPV testing has high sensitivity and negative predictive value, but low specificity, which can easily lead to false positives, resulting in high referral rates and high costs. Therefore, the present invention aims to explore the composition of reproductive tract microorganisms associated with high-grade cervical lesions, assist the current cervical cancer screening methods (HPV testing and cervical cytology), improve the accuracy of detection, and reduce unnecessary invasive testing and excessive treatment.

[0005] In view of this, the present invention analyzes the variation abundance of vaginal flora in cervical cancer based on 16S sequencing data, screens and identifies microbial markers in people at high risk of cervical cancer, and constructs a risk prediction model, aiming to solve the problems existing in current cervical cancer screening methods and improve the accuracy of early screening of cervical cancer. Summary of the invention

[0006] In order to improve the efficiency of early screening of cervical lesions and solve the problem of late onset, the relative abundance of reproductive tract microbial markers is obtained based on 16S sequencing data by collecting cervical scraping cells of samples, a random forest model is established, and the threshold of output risk value is obtained. The relative abundance of 11 reproductive tract microbial markers that have been screened is obtained in the test set, and the established random forest model is input to obtain the probability value of cervical lesion samples being high-risk lesions. The present invention provides a biological model for predicting the risk of cervical cancer, which can help discover potential abnormal cell changes in the early stage and help clinicians strengthen the effective diagnosis of the degree of cervical lesions in patients.

[0007] To achieve the above object, the technical solution of the present invention is as follows: In a first aspect, in an embodiment of the present invention, a method for constructing a cervical cancer microbial marker risk prediction model is provided, and the steps of constructing the model are as follows: S1-1: Using 16S sequencing, respectively detect the YZ_LSIL and HSIL_CC samples in the patients to obtain the bacterial relative abundance data of the microbial markers described in the first aspect; S1-2: The data obtained in step S1-1 is randomly divided into a training set and a test set according to a ratio of 8:2, the training set machine learning model is used, and GridsearchCV is used to adjust and optimize the parameters; S1-3: Repeat the S1-2 model parameter adjustment and optimization steps 200 times to obtain the sensitivity and specificity sequence sets, draw the ROC curve, calculate the average ROC-AUC of the test set, and select the model with the highest ROC-AUC score as the optimal model; S1-4: Use the optimal model selected in S1-3 to build a cervical cancer risk prediction model, select the best 11 species of bacteria to input into the optimal model, test and save the trained model.

[0008] Preferably, the method for determining the cervical cancer microbial markers includes any one or a combination of at least two of 16S sequencing, metagenomic sequencing or qPCR quantitative detection.

[0009] Preferably, the machine learning model in step S1-2 includes Logistic regression, support vector machine, random forest and Xgboost, preferably random forest.

[0010] In the embodiment of the present invention, the relative abundance of a certain type of bacteria is used as a prediction index, and the measurement means of the relative abundance of bacteria are various. Therefore, it can break the limitation of specific test equipment and experimental skills and has good versatility.

[0011] Second aspect, in the embodiments of the present invention, a method for selecting bacteria at the species level is provided. The method for selecting bacteria at the species level includes the following steps: S2-1: Obtain the importance scores of variable features using the above-mentioned preferred random forest model, gradually adjust the number of bacterial variables according to the high and low importance scores, and obtain the variables required when the ROC-AUC is optimal. When the input feature variables are the relative abundances of 11 species-level bacteria, the ROC-AUC value is the largest; S2-2: Model testing. Use the relative abundance data of 11 species-level bacteria corresponding to the training set samples to construct a random forest model, and obtain the optimal threshold a for the output risk value; S2-3: Model storage. Save the trained model for subsequent prediction of cervical cancer risk.

[0012] Preferably, for the optimal threshold of the output risk value in the above S2-2, the determination method of the sample output value is as follows: (1) Output risk value < a, determined as YZ_LSIL; (2) Output risk value ≥ a, determined as HSIL_CC, and it is recommended to perform a clinical colposcopy.

[0013] Preferably, the present invention provides a variety of cervical cancer microbial markers. The 11 species-level microbial markers in S2-2 include Porphyromonas asaccharolytica ( Porphyromonas asaccharolytica ), Lactobacillus crispatus ( Lactobacillus crispatus ), Klebsiella pneumoniae ( Klebsiella pneumoniae ), Gemella haemolysans ( Gemella haemolysans ), Porphyromonas sp.-type-1, Megasphaera sp.-type-1, Lactobacillus jensenii ( Lactobacillus jensenii ), Lactobacillus vaginalis ( Lactobacillus vaginalis ), Mycoplasma girerdii , JH815513 and Lactobacillus gasseri ( Lactobacillus gasseri ).

[0014] Preferably, the cervical cancer microbial markers have good sensitivity and specificity, and are helpful for auxiliary diagnosis or early warning of the risk of cervical cancer.

[0015] Preferably, the auxiliary diagnosis includes distinguishing HSIL_CC patient samples from the test samples.

[0016] Further preferably, the nucleic acid detection reagent for the microbial marker is a 16S sequencing detection reagent.

[0017] Further preferably, the 16S sequencing detection reagent for the microbial marker is a sequencing library detection reagent for the V4 region of the bacterial 16S rRNA gene.

[0018] In the embodiments of the present invention, the precancerous lesions of cervical cancer are clinically divided into three levels: CIN1, CIN2, and CIN3, and the classification is based on the proportion of cells with abnormal nuclear mitotic figures in the cervical squamous epithelium. The HSIL described in the present invention includes two types of lesions, CIN2 and CIN3, and the LSIL includes CIN1. YZ refers to inflammation of the cervix, including inflammation of the cervicovaginal part (i.e., the part of the lower end of the cervix leading to the vagina) and inflammation of the cervical mucosa. In addition, the high-risk samples described in the present invention refer to samples with test results of HSIL or CC, and the low-risk samples refer to samples with test results of LSIL or YZ.

[0019] In an embodiment of the present invention, cross validation is a statistical method or resampling procedure used to evaluate the likelihood of a machine learning model on a limited sample of data.

[0020] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention collects cervical scraping cell samples and constructs a model based on the 16S sequencing data set. Compared with cervical cytology and HPV testing, it is more convenient and faster, and the cost of 16S sequencing technology is relatively low.

[0021] 2. The present invention provides a variety of cervical cancer microbial markers that can be used for the prediction and auxiliary diagnosis of high-risk and low-risk cervical lesions. The detection has high sensitivity and specificity, which helps to reduce the incidence of cervical cancer.

[0022] 3. The present invention constructs a cervical cancer risk prediction model based on the selected microbial markers, which has high performance and helps to improve the accuracy of cervical cancer risk prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of model construction in Example 1 of the present invention.

[0024] Figure 2 This is the feature importance ranking output by the random forest model in Example 2 of the present invention.

[0025] Figure 3 This is a ROC curve diagram of the model validation in Example 3 of the present invention. DETAILED DESCRIPTION

[0026] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0027] The main abbreviations and key terms involved in the present invention are explained as follows: 16S rRNA: A type of rRNA involved in the production of the small subunit of prokaryotic ribosomes.

[0028] 16S sequencing: This sequencing is based on the 16S rRNA gene. It obtains the partial region or full-length sequence of the 16S rRNA gene through PCR amplification; AUC:Area Under the Curve Area under the curve; ROC curve: Receiver Operating Characteristic Curve Receiver operating characteristic curve; HPV: Human Papillomavirus; CIN1: Cervical Intraepithelial Neoplasia 1, mild atypical hyperplasia of cervical squamous epithelium; CIN2: Cervical Intraepithelial Neoplasia 2, moderate atypical hyperplasia of cervical squamous epithelium; CIN3: Cervical Intraepithelial Neoplasia 3, severe atypical hyperplasia and carcinoma in situ of cervical squamous epithelium; HSIL: High Grade Squamous Intraepithelial Lesion, high grade squamous intraepithelial lesion; LSIL: Low Grade Squamous Intraepithelial Lesion, low grade squamous intraepithelial lesion; YZ: cervicitis; Logistic regression: Logistic regression is a generalized linear regression analysis model, which is often used in data mining, automatic disease diagnosis, economic forecasting and other fields.

[0029] Xgboost: eXtreme Gradient Boosting, an optimized distributed gradient boosting library that is efficient, flexible and portable.

[0030] GridSearch CV: GridSearch and CV, namely grid search and cross validation, are used to systematically traverse multiple parameter combinations and determine the best effect parameters through cross validation.

[0031] Example 1 Construction of a cervical cancer risk prediction model using microbial markers

[0032] 1. Sample information A total of 334 cervical cancer lesion samples were collected from Anhui Provincial First People's Hospital, including 137 patients with low-grade lesions and 197 patients with high-grade lesions.

[0033] Model construction (see Figure 1 ) S1-1: 16S sequencing was used to detect the collected YZ_LSIL and HSIL_CC samples, and the relative abundance values ​​of 396 bacterial species were obtained; S1-2: The data obtained in step S1-1 is randomly divided into a training set and a test set according to a ratio of 8:2. The training set is sequentially input into Logistic regression, support vector machine, random forest, and Xgboost, and GridsearchCV is used to adjust and optimize the parameters; S1-3: Repeat the above steps 200 times to obtain the sensitivity and specificity sequence sets, draw the ROC curve, and calculate the average ROC-AUC of the test set. The average ROC-AUCs of Logistic, Support Vector Machine, Random Forest, and Xgboost were 0.762, 0.753, 0.893, and 0.876, respectively. The random forest model with the highest ROC-AUC score was selected as the optimal model (see Table 1).

[0034] Table 1. Comparison of model performance of different algorithms algorithm AUC Logistic 0.762±0.051 Support Vector Machine 0.753±0.058 Random Forest 0.893±0.035 Xgboost 0.876±0.039 S1-4: Use the optimal random forest algorithm selected in S1-3 to build a cervical cancer risk prediction model, select the optimal 11 species-level bacteria to input into the random forest model, test and save the trained model.

[0035] Example 2 Selection of bacteria at the species level

[0036] S2-1: Use the random forest model to obtain the importance score of variable features, and gradually adjust the number of bacterial variables according to the importance score to obtain the variables required for the optimal ROC-AUC (see Figure 2 ), the ROC-AUC value is the largest when the input feature variable is the relative abundance of bacteria at the 11 specific species levels; S2-2: Model testing. The random forest model was constructed using the bacterial relative abundance data at the 11 species level corresponding to the training set samples, and the optimal threshold of the output risk value was obtained as 0.29; S2-3: Model storage. Save the trained model for subsequent prediction of cervical cancer risk.

[0037] Preferably, according to the above step S2-2, the optimal threshold value of the risk value is outputted as 0.29, and the sample category is determined. The determination method of the output value is as follows: (1) Output risk value < 0.29, judged as YZ_LSIL; (2) If the output risk value is ≥ 0.29, it is judged as HSIL_CC and clinical colposcopy is recommended.

[0038] In an embodiment of the present invention, step S2-1 will generate different ROC-AUC values ​​according to the data size and variable combination. In an embodiment of the present invention, the inventor selected the relative abundance of 11 species of bacteria that are the optimal input variables for matching with the model, that is, using the 11 species of bacteria described in the present invention can obtain higher prediction performance.

[0039] Example 3 Model Validation

[0040] (1) Samples were collected from 8 patients with high-grade lesions and 13 patients with low-grade lesions, and 16S sequencing was performed on the 21 samples. The relative abundance of 11 specific species of bacteria was selected and the relative abundance data were input into the model; (2) Output predicted risk value: Input the sequencing data obtained in the experiment into the established random forest model to obtain the output risk value. The result determination method is as follows: the output risk value is the probability between 0 (control) and 1 (cervical cancer patients), and the optimal threshold of the output risk value obtained is 0.29. Samples with an output risk value greater than or equal to 0.29 probability value are determined as HSIL_CC, and samples with a probability value less than 0.29 probability value are determined as YZ_LSIL.

[0041] Table 2. Actual disease status of 21 subjects and risk values ​​output by the prediction model Sample No. Sample Type Model output risk value 1 YZ_LSIL 0.206197 2 YZ_LSIL 0.267742 3 YZ_LSIL 0.108255 4 YZ_LSIL 0.289961 5 YZ_LSIL 0.115925 6 YZ_LSIL 0.265199 7 YZ_LSIL 0.143235 8 YZ_LSIL 0.583998 9 YZ_LSIL 0.129725 10 YZ_LSIL 0.195822 11 YZ_LSIL 0.101491 12 YZ_LSIL 0.25737 13 YZ_LSIL 0.042597 14 HSIL_CC 0.425796 15 HSIL_CC 0.425796 16 HSIL_CC 0.35092 17 HSIL_CC 0.35092 18 HSIL_CC 0.187282 19 HSIL_CC 0.296521 20 HSIL_CC 0.481914 21 HSIL_CC 0.481914 As can be seen from Table 2, the cervical cancer microbial markers provided by the present invention have high prediction accuracy and can effectively distinguish YZ_LSIL and HSIL_CC in multiple samples.

[0042] In the embodiment of the present invention, the 11 selected optimal species-level bacteria were input into the random forest model to obtain the risk prediction value and verify it. The ROC curve effect of the random forest model (see Figure 3 ). The ROC-AUC of the model prediction is 0.865, which has good performance.

[0043] It should be noted that the above content only illustrates the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications all fall within the protection scope of the claims of the present invention.

Claims

1. A cervical cancer microbial marker, characterized in that: The microbial markers include 11 species-level bacteria, namely: Porphyromonas aglycolyticus ( Porphyromonas asaccharolytica )、Lactobacillus crispatus( Lactobacillus crispatus ), Klebsiella pneumoniae ( Klebsiella pneumoniae ), Gemini cocci ( Gemella haemolysans )、Porphyromonas type 1( Porphyromonas type 1 )、Megalosphaeria type 1( Megasphaera type 1 ), Lactobacillus jensenii ( Lactobacillus jensenii )、Vaginal Lactobacillus( Lactobacillus vaginalis )、 Mycoplasma girerdii , JH815513 and Lactobacillus gasseri ( Lactobacillus gasseri ).

2. A model for predicting cervical cancer risk, characterized in that: The input variable of the model is the relative abundance of the microbial markers described in Claim 1 or Claim 2.

3. The relative abundance of microbial markers of the input variables according to claim 3, characterized in that The method for measuring the microbial markers is any one of 16S sequencing, metagenomic sequencing or qPCR quantitative detection, preferably 16S sequencing.

4. A method for constructing a cervical cancer microbial marker risk prediction model, characterized in that: The model construction includes the following steps: S1-1: Using 16S sequencing, detect the collected YZ_LSIL and HSIL_CC samples respectively, and obtain the bacterial relative abundance data of the microbial markers described in the first aspect. S1-2: Randomly divide the data obtained in step S1-1 into a training set and a test set according to 8:

2. The training set is input into the machine learning model, and GridsearchCV is used to adjust and optimize the parameters. S1-3: Repeat the model parameter adjustment and optimization steps in S1-2 for 200 times to obtain a sensitivity and specificity sequence set, draw an ROC curve, calculate the area under the curve (ROC-AUC) of the average receiver operating characteristic curve of the test set, and select the model with the highest ROC-AUC score as the optimal model. S1-4: Use the optimal model selected in S1-3 to construct a cervical cancer risk prediction model, select the optimal 11 bacterial species at the species level and input them into the optimal model, and test and save the trained model.

5. The method for constructing a cervical cancer microbial marker risk prediction model according to claim 5, characterized in that: In step S1-2, the machine learning models include Logistic regression, support vector machine, random forest and Xgboost, preferably random forest.

6. The method for selecting bacteria at the specific species level according to S1-4 in claim 5, characterized in that: The method for selecting the bacterial species at the species level includes the following steps: S2-1: Use the preferred random forest model described in Claim 5 to obtain the importance scores of the variable features, gradually adjust the number of bacterial variables according to the high and low importance scores, and obtain the variables required when the ROC-AUC is optimal. When the input feature variables are the relative abundances of 11 bacterial species at the species level, the ROC-AUC value is the largest. S2-2: Model testing. Use the relative abundance data of 11 bacterial species at the species level corresponding to the training set samples to construct a random forest model, and obtain the optimal threshold a for the output risk value. S2-3: Model storage. Save the trained model for subsequent prediction of cervical cancer risk.

7. According to claim 6, S2-2 outputs the optimal risk value threshold a, characterized in that: The determination method of the output value is as follows: Output risk value < a, determined as YZ_LSIL; Output risk value ≥ a, determined as HSIL_CC, and clinical colposcopy is recommended.

Images