A biomarker combination for auxiliary judgment of cervical lesions and application thereof

Abstract

The present application relates to the technical field of cervical lesion molecular judgment, in particular to a biomarker combination for auxiliary judgment of cervical lesion, which comprises MIF, SYNGR1 and TOPBP1, the biomarker combination is used for auxiliary identification of cervical high-grade squamous intraepithelial lesion and cervical cancer patient, the biomarker combination is used for distinguishing low-risk lesion group and high-risk lesion group, wherein the low-risk lesion comprises normal cervical tissue and low-grade squamous intraepithelial lesion, and the high-risk lesion comprises high-grade squamous intraepithelial lesion and cervical cancer tissue.The biomarker combination for auxiliary judgment of cervical lesion and application, for the first time, establish an auxiliary judgment model based on the combined expression mode of MIF, SYNGR1 and TOPBP1 three genes, and for the first time, find that the combined expression of MIF, SYNGR1 and TOPBP1 can be used as a molecular marker combination of cervical high-grade lesion and cervical cancer, and can be used for distinguishing low-risk lesion group and high-risk lesion group.

Description

Technical Field

[0001] This invention relates to the field of molecular diagnostic technology for cervical lesions, specifically to a combination of biomarkers for the auxiliary diagnosis of cervical lesions and their application. Background Technology

[0002] Cervical cancer is the fourth most common malignant tumor among women worldwide, seriously threatening the health and lives of many women. According to statistics from 2020, there were approximately 604,000 new cases of cervical cancer and 342,000 deaths globally. In my country, the incidence and mortality rates of cervical cancer remain high, with approximately 110,000 new cases and 59,000 deaths annually, showing an upward trend compared to 2018. Although the mortality rate of cervical cancer has decreased in some regions due to the global promotion of human papillomavirus (HPV) vaccination and the three-step cervical cancer screening program, it remains a leading cause of death for women in many developing countries and areas with relatively scarce medical resources.

[0003] Currently, the most widely used cervical cancer screening methods in clinical practice mainly rely on cytological examinations (such as TCT) and HPV testing. However, both methods have certain limitations in practical applications: while cytological examinations have good specificity, the interpretation of results is highly subjective, and the sensitivity is relatively low, making it easy to miss diagnoses; while HPV testing has high sensitivity, its specificity is insufficient, and due to the common occurrence of transient infections, it is prone to false positive results, leading to unnecessary colposcopy referrals and overtreatment. Traditional histopathological diagnosis, as the gold standard for diagnosis, is highly accurate, but it is an invasive examination, relies on the experience of pathologists, and is difficult to meet the needs of large-scale population screening and precise stratified management. Therefore, developing a more objective, accurate, and easily scalable molecular diagnostic method is of significant clinical importance.

[0004] The development of cervical cancer is a continuous evolutionary process, typically progressing from "normal cervical epithelium → low-grade squamous intraepithelial lesion (LSIL) → high-grade squamous intraepithelial lesion (HSIL) → invasive carcinoma." Among these, high-grade squamous intraepithelial lesion (HSIL) is considered the most important precancerous stage and a critical window for clinical intervention and intervention to prevent the development of cervical cancer. Therefore, accurately identifying HSIL and cervical cancer patients, and distinguishing them from low-risk lesions, is of core value for improving early screening efficiency, optimizing clinical decision-making, and improving patient prognosis.

[0005] In recent years, biomarkers have received considerable attention in the auxiliary diagnosis of cervical cancer. Currently, the combined detection of CDKN2A (p16) and MKI67 (Ki-67), commonly used in clinical practice, while aiding pathological diagnosis to some extent, still suffers from insufficient sensitivity and specificity in differentiating low-risk lesions (including normal tissue and LSIL) from high-risk lesions (HSIL and cervical cancer), failing to fully meet the needs of precision medicine. Furthermore, most existing biomarker detections are based on protein levels, heavily reliant on the subjective interpretation of pathologists, lacking standardized and objective judgment models based on the transcriptome level, making it difficult to achieve stable and reliable risk stratification across different detection platforms.

[0006] In summary, there is an urgent need to develop a combination of MIF-SYNGR1-TOPBP1 three-gene biomarkers for the auxiliary diagnosis of high-grade squamous intraepithelial lesions (HSIL) and cervical cancer, and to address the problems of insufficient sensitivity, insufficient specificity, and poor cross-platform stability of existing cervical lesion auxiliary diagnostic markers in distinguishing between low-risk and high-risk lesions. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a combination of biomarkers for the auxiliary diagnosis of cervical lesions and their application, which can accurately distinguish between low-risk and high-risk lesions at the transcriptomic level, has high sensitivity and specificity, and exhibits good cross-platform stability.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a combination of biomarkers for the auxiliary diagnosis of cervical lesions, comprising: MIF, SYNGR1 and TOPBP1.

[0009] Furthermore, the biomarker combination is used to assist in the identification of patients with high-grade squamous intraepithelial lesions and cervical cancer.

[0010] Furthermore, the biomarker combination is used to distinguish between low-risk lesion groups and high-risk lesion groups, wherein the low-risk lesions include normal cervical tissue and low-grade squamous intraepithelial lesions, and the high-risk lesions include high-grade squamous intraepithelial lesions and cervical cancer tissue.

[0011] A detection kit prepared based on the combination of said biomarkers, comprising:

[0012] Primers and / or probes for detecting MIF expression levels;

[0013] Primers and / or probes for detecting SYNGR1 expression levels;

[0014] Primers and / or probes for detecting TOPBP1 expression levels;

[0015] Internal reference gene detection components:

[0016] And the amplification reaction system.

[0017] An application of the aforementioned biomarker combination in the preparation of an auxiliary diagnostic product for cervical lesions, wherein the auxiliary diagnostic product is a diagnostic kit, gene chip, or sequencing library.

[0018] A method for assisting in the diagnosis of cervical lesions based on the aforementioned biomarker combination includes the following steps:

[0019] Obtain expression data of MIF, SYNGR1, and TOPBP1 in the sample to be tested;

[0020] The expressed data are standardized.

[0021] The standardized representation data is input into a pre-built classification model;

[0022] The model output results determine whether the sample belongs to the low-risk lesion group or the high-risk lesion group.

[0023] Furthermore, the method for constructing the classification model includes the following steps:

[0024] Single-cell transcriptome data of the training samples were obtained, and pseudo-bulk RNA-seq expression matrix was constructed using the AggregateExpression function;

[0025] Logarithmic transformation was performed on the pseudo-bulk RNA-seq expression matrix, and then centered and normalized based on the training set.

[0026] Using the expression values ​​of MIF, SYNGR1, and TOPBP1 as input variables and sample grouping as the outcome variable, a binary logistic regression model, i.e., a classification model, is constructed using machine learning algorithms.

[0027] Furthermore, the machine learning algorithm includes one or more of glmnet-ridge, ridge logistic regression, lasso logistic regression, elastic-net logistic regression, random forest, support vector machine, or xgboost.

[0028] A system for assisting in the diagnosis of cervical lesions, comprising:

[0029] The data acquisition module is used to acquire data on MIF, SYNGR1, and TOPBP1 in the sample to be tested;

[0030] The data processing module is used to standardize the expressed data;

[0031] The risk assessment module is used to calculate the category to which a sample belongs based on a preset classification model;

[0032] The result output module is used to output auxiliary judgment results.

[0033] A computer-readable storage medium having a computer program stored thereon, the computer program implementing the auxiliary judgment method when executed by a processor.

[0034] Compared with the prior art, the technical solution of this application has the following beneficial effects:

[0035] 1. This invention establishes for the first time an auxiliary judgment model based on the joint expression pattern of three genes, MIF, SYNGR1 and TOPBP1, and discovers for the first time that the joint expression of MIF, SYNGR1 and TOPBP1 can be used as a molecular marker combination for high-grade cervical lesions and cervical cancer, which can be used to distinguish between low-risk lesion groups and high-risk lesion groups.

[0036] 2. The present invention has excellent auxiliary judgment performance. The three-gene model constructed by the present invention obtained an AUC value of 0.923 in the training set, with a sensitivity of 0.910 and a specificity of 0.723. It can effectively distinguish the low-risk lesion group, namely Normal, Para-tumor and LSIL, and the high-risk lesion group, namely HSIL and cervical cancer, showing good classification ability.

[0037] 3. This invention has good cross-platform stability and generalization ability, and maintains stable performance in TCGA-CESC and multiple independent external validation datasets. It is applicable to RNA sequencing platforms and expression microarray platforms, demonstrating its good cross-platform applicability and clinical application potential.

[0038] 4. The judgment performance of the present invention is superior to that of existing biomarker combinations. Specifically, compared with single-gene models, dual-gene models and traditional CDKN2A+MKI67 biomarker combinations, the present invention can achieve better auxiliary judgment performance using only three genes, reducing detection complexity and detection cost while ensuring accuracy.

[0039] 5. This invention has good clinical application value and can be used for the auxiliary judgment, early screening and risk stratification of high-grade cervical lesions and cervical cancer. It helps to improve the identification efficiency of high-risk patients and provides objective and reliable molecular evidence for clinical diagnosis and treatment decisions. Attached Figure Description

[0040] Figure 1The ROC curves of the MIF-SYNGR1-TOPBP1 three-gene model and control model of this invention in the training cohort are used to compare the classification performance of single-gene models, double-gene models, triple-gene models and traditional CDKN2A+MKI67 biomarker combinations.

[0041] Figure 2 This is the ROC curve of the MIF-SYNGR1-TOPBP1 three-gene model and control model of this invention on the TCGA-CESC independent external validation set.

[0042] Figure 3 This is the ROC curve of the MIF-SYNGR1-TOPBP1 three-gene model and control model of this invention in the GSE113942 independent external validation set.

[0043] Figure 4 This is the ROC curve of the MIF-SYNGR1-TOPBP1 three-gene model and control model of this invention on the GSE9750 independent external validation set.

[0044] Figure 5 This is the ROC curve of the MIF-SYNGR1-TOPBP1 three-gene model and control model of this invention in the GSE27678 independent external validation set.

[0045] Figure 6 This is the ROC curve of the MIF-SYNGR1-TOPBP1 three-gene model and control model of this invention in the GSE63514 independent external validation set.

[0046] Figure 7 This is the ROC curve of the MIF-SYNGR1-TOPBP1 three-gene model of the present invention in the clinical sample validation set. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Please see Figure 1-7 This embodiment includes a combination of biomarkers for assisting in the diagnosis of cervical lesions, including MIF, SYNGR1, and TOPBP1, which are used to help identify patients with high-grade squamous intraepithelial lesions (HSIL) and cervical cancer, and to distinguish between low-risk and high-risk lesion groups.

[0049] The low-risk lesion group includes Normal, Para-tumor, and LSIL samples, namely normal cervical tissue and low-grade squamous intraepithelial lesion; the high-risk lesion group includes HSIL and Tumor samples, namely high-grade squamous intraepithelial lesion and cervical cancer tissue.

[0050] To facilitate convenient and standardized detection of the above-mentioned biomarker combinations, this invention also provides a cervical lesion auxiliary diagnostic kit.

[0051] Preferably, the kit includes detection reagents for detecting the expression levels of each biomarker in the biomarker combination.

[0052] More preferably, the detection reagent comprises:

[0053] Primers and / or probes used to detect MIF expression levels;

[0054] Primers and / or probes used to detect SYNGR1 expression levels;

[0055] Primers and / or probes used to detect TOPBP1 expression levels;

[0056] Internal reference gene detection components;

[0057] Amplification reaction system.

[0058] The present invention also provides an application of the combination of the aforementioned biomarkers in the preparation of an auxiliary diagnostic product for cervical lesions, wherein the auxiliary diagnostic product is a test kit, gene chip, or sequencing library.

[0059] The biomarker combination provided in this embodiment can distinguish between low-risk and high-risk cervical lesions. Based on this, the present invention also provides a method for assessing the risk of cervical lesions based on the biomarker combination, including the following steps:

[0060] S1. Obtain the expression data of MIF, SYNGR1 and TOPBP1 in the sample to be tested;

[0061] S2. Standardize the expressed data;

[0062] S3. Input the standardized expression data into the pre-built classification model;

[0063] S4. Determine whether the sample to be tested belongs to the low-risk lesion group or the high-risk lesion group based on the model output results.

[0064] As a preferred method, the construction of the classification model includes the following steps:

[0065] S1. Obtain single-cell transcriptome data of the training samples and construct a pseudo-bulk RNA-seq expression matrix using the AggregateExpression function;

[0066] S2. Perform a log2(x+1) transformation on the representation matrix, and center and standardize it based on the training set;

[0067] S3. Using the expression values ​​of MIF, SYNGR1, and TOPBP1 as input variables and sample grouping as the outcome variable, a binary logistic regression model, i.e., a classification model, is constructed using machine learning algorithms.

[0068] The machine learning algorithms include one or more of glmnet-ridge, ridge logistic regression, lassologistic regression, elastic-net logistic regression, random forest, supportvector machine, or xgboost.

[0069] This invention constructs a pseudo-bulk expression matrix based on single-cell transcriptome data and combines various machine learning algorithms for feature selection and model optimization, ultimately obtaining the optimal three-gene combination. In the training set, this three-gene model achieves an AUC of 0.923, a sensitivity of 0.910, and a specificity of 0.723, demonstrating good discriminative ability. It exhibits cross-platform stability and generalization ability in multiple independent external validation sets such as TCGA, showcasing the performance improvement of the three-gene model compared to single-gene, dual-gene, and traditional CDKN2A+MKI67 biomarker combinations. The three-gene combination described in this invention achieves superior diagnostic performance with a smaller number of feature genes and can be used for the auxiliary diagnosis, early screening, and risk stratification of HSIL and cervical cancer, providing objective and reliable molecular evidence for precision clinical diagnosis and treatment.

[0070] In one specific implementation, regarding the source of training data and sample grouping, this project uses single-cell transcriptome data containing 143 cervical tissue samples as the training set. First, the single-cell data is aggregated at the sample level, and a pseudo-bulk RNA-seq expression matrix is ​​generated for each sample. The AggregateExpression function of Seurat is used to aggregate the expression levels of cells from the same sample source according to the gene dimension, resulting in a "gene × sample" sample-level expression matrix, which is used for subsequent classification model construction.

[0071] Based on pathological or clinical grouping information, the samples were divided into two categories: Normal, Para-tumor, and LSIL were classified into the low-risk / non-tumor progression group, referred to as the Normal group; HSIL and Tumor were classified into the high-risk / tumor progression group, referred to as the Tumor group. In the training set, there were 65 cases in the Normal group and 78 cases in the Tumor group, for a total of 143 cases.

[0072] For expression matrix preprocessing, specifically, the expression matrices of both the training and validation sets are preprocessed uniformly. For RNA-seq or pseudo-bulk RNA-seq expression data, a logarithmic transformation is performed; for microarray expression matrices already on a logarithmic scale or containing values ​​less than or equal to -1, the logarithmic transformation is not repeated. For data requiring logarithmic transformation, the following method is used:

[0073]

[0074] in, This represents the original expression value of gene g in sample i.

[0075] Then, each gene is standardized using the mean and standard deviation of the training set:

[0076]

[0077] The validation set is calculated using the training set. and The standardized parameters are not re-estimated in the validation set to avoid validation information entering the model training process.

[0078] For the validation set using Ensembl IDs, the Ensembl IDs are converted into gene symbols. For cases where the same gene symbol corresponds to multiple probes, multiple Ensembl transcripts, or multiple repeat lines, they are first uniformly mapped to the same gene symbol. Then, the arithmetic mean of the expression values ​​of these repeat lines is calculated for each sample, and this is taken as the final expression level of the gene in the corresponding sample. Only target genes that can be matched in both the training and validation sets are retained for model prediction.

[0079] Specifically, in the process of identifying core candidate genes and constructing feature combinations, feature selection is not based solely on statistical differences or machine learning algorithm results during candidate gene screening. Instead, it combines the results of previous single-cell sequencing analysis, cell communication analysis, and molecular mechanism experiments to determine MIF as the core candidate biomarker, which is then consistently incorporated into the subsequent feature combination construction and classification model training.

[0080] Previously, we constructed and analyzed single-cell transcriptome data from a K14 mouse model simulating HPV integration-mediated cervical cancer development. This model covers multiple key stages of cervical cancer development, including the HPV integration-negative stage, the HPV integration-positive stage, low-grade squamous intraepithelial lesion (LSIL), high-grade squamous intraepithelial lesion (HSIL), and the tumor stage. By performing single-cell RNA sequencing and cell type annotation on cervical tissue samples from the above different stages, we further identified different epithelial cell subpopulations in the epithelial cell population, such as normal keratinocytes, precancerous epithelial cells, and tumor cells. This provides a foundation for understanding the epithelial cell state transitions and their microenvironment interactions during the development of cervical cancer.

[0081] Based on the single-cell data mentioned above, further cell communication analysis revealed that the interaction between epithelial cells and fibroblasts was the most significant during the progression of cervical lesions, suggesting that epithelial-mesenchymal cell interactions may be involved in the progression of cervical cancer and the formation of high-grade lesions. Specifically, the communication strength of the MIF-ACKR3 ligand-receptor pair was significantly enhanced in the HSIL and Tumor stages, indicating that MIF-mediated signaling between epithelial cells and fibroblasts is closely related to the formation of high-grade cervical lesions and cervical cancer.

[0082] Experimental results showed that the MIF protein secreted by epithelial cells can bind to the ACKR3 receptor on the surface of fibroblasts and activate the downstream NF-κB signaling pathway in fibroblasts, thereby promoting the formation of the tumor-associated microenvironment and the progression of cervical lesions. Simultaneously, the expression of the MIF gene in epithelial cells is regulated by the HPV integration-mediated E2F1 transcription factor, indicating that MIF is a key link between HPV integration events, abnormal transcriptional regulation in epithelial cells, and abnormal activation of epithelial-mesenchymal cell communication.

[0083] Based on the above findings, MIF not only exhibits stage-related expression characteristics during the development and progression of cervical cancer, but is also closely related to enhanced epithelial-fibroblast communication in the HSIL and Tumor stages, activation of the NF-κB signaling pathway, and transcriptional regulation related to HPV integration. Therefore, this invention designates MIF as a core candidate gene with clear biological evidence and disease progression indicative significance, and consistently incorporates it into all candidate biomarker combinations and machine learning classification models to improve the model's ability to capture molecular alterations related to high-grade cervical lesions and cervical cancer, as well as its biological interpretability.

[0084] In addition to MIF, non-MIF candidate genes related to cervical lesion progression, malignant transformation of epithelial cells, and model classification performance were further screened from the candidate gene pool. While ensuring that MIF was consistently included, 1 to 9 non-MIF genes were selected from the candidate gene pool to construct a MIF central composite model containing 2 to 10 genes.

[0085] For machine learning model training, multiple supervised learning algorithms are used to construct a binary classification model for each candidate gene combination. Model training uses the training set group labels as the outcome variable and the candidate gene expression levels as the input variable. Candidate algorithms include ridge logistic regression, elastic-net logistic regression, lassologistic regression, ordinary logistic regression, k-nearest neighbors, linear discriminant analysis, naive Bayes, random forest, XGBoost, and radial-kernel support vector machine. For the logistic model, the predicted probability of a sample belonging to the high-risk group is:

[0086]

[0087] in, This represents the probability that sample i is classified as a high-risk group; The intercept; Let be the regression coefficient of the j-th gene; This represents the standardized gene expression value.

[0088] The glmnet model estimates parameters by minimizing the following formula:

[0089]

[0090] in ; Ridge return to take Elastic networks take LASSO Penalty parameters Cross-validation is selected within the training set to optimize cross-validation performance. Since the expression values ​​have already been standardized using the training set parameters, glmnet does not perform the standardization again internally. The K-nearest neighbor model uses k=9 in this training.

[0091] For all models, the following metrics were recorded: training set cross-validation AUC, training set AUC, validation set AUC, accuracy, sensitivity, specificity, F1 score, balanced accuracy, and MCC. The classification threshold was determined based on the Youden exponent from the training set ROC curve.

[0092]

[0093] Choose to Threshold for reaching the maximum value Furthermore, this threshold is consistently used for class determination on the validation set to avoid bias caused by re-optimizing the threshold on the validation set.

[0094] Regarding the final determination of the three-gene model, the following factors were considered when selecting the final model from the candidate models: First, the AUC performance of the model on the training set and the independent validation set; Second, the AUC decrease between the training set and the validation set; Third, the stability in different platforms and cohorts; Fourth, the number of features and the feasibility of clinical testing.

[0095] While some four- or more-gene models can achieve high AUCs, from the perspectives of simplicity, conversion cost, and overfitting risk control, the three-gene model, which maintains high predictive performance with fewer features, is preferred. The three-gene model consisting of MIF, SYNGR1, and TOPBP1 was ultimately selected. This model achieved a cross-validation AUC of 0.9281 on the training set and a training set AUC of 0.9233, indicating good discriminative ability on the training set.

[0096] The final model was built using the glmnet-ridge framework. The model input consisted of the standardized expression values ​​of MIF, SYNGR1, and TOPBP1 in the samples, and the output was the predicted probability that a sample belonged to the high-risk / tumor progression group. Based on a threshold determined from the training set, the predicted probability was converted into a negative or positive classification result.

[0097] To evaluate the model's cross-platform generalization ability, multiple independent, publicly available, or proprietary cervical-related RNA expression datasets were collected as external validation sets for external validation. These datasets included RNA-seq and expression array datasets. The RNA-seq validation set included GSE113942, GSE192804, GSE223804, GSE87410, the TCGA-CESC cohort, and the WPP cohort; the array validation set included GSE27678, GSE63514, GSE7410, GSE7803, and GSE9750.

[0098] The external validation set used a uniform binary classification standard: Normal, Para-tumor, LSIL, CIN1, and CIN2 were classified as low-risk / non-tumor progression group; HSIL, CIN3, and Tumor were classified as high-risk / tumor progression group. For each validation set, AUC, accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1 score, balanced accuracy, and MCC were calculated.

[0099] As for the comparison with control models, in order to demonstrate the predictive gain and stability of the MIF+SYNGR1+TOPBP1 three-gene model, the following control models were further constructed and compared: SYNGR1 single-gene model, TOPBP1 single-gene model, CDKN2A+MKI67 double-gene model (traditional biomarker combination), MIF+SYNGR1 double-gene model, MIF+TOPBP1 double-gene model, and MIF+SYNGR1+TOPBP1 three-gene model.

[0100] To evaluate the clinical applicability of the MIF-SYNGR1-TOPBP1 trigene model, cervical exfoliated cell samples were collected from 316 subjects in an outpatient setting using a cervical brush sampler. Experienced clinicians categorized the samples into high-risk / tumor progression and low-risk / non-tumor progression groups, with 136 high-risk samples and 180 low-risk samples. Total RNA was extracted from the samples, and cDNA was synthesized using a reverse transcription kit. The expression levels of MIF, SYNGR1, and TOPBP1 were detected using real-time quantitative PCR (RT-qPCR). The model was applied to predict outcomes in the 316 samples, with the physician-confirmed high-risk or low-risk classification used as the reference outcome. AUC values, sensitivity, and specificity were calculated.

[0101] The models described above were compared using the same training set, preprocessing procedure, standardized parameters, and evaluation metrics. Predicted probabilities, classification results, and ROC curves were generated for each model on the training set and all external validation sets. By comparing the AUC distribution, sensitivity / specificity balance, and cross-platform performance of different models across multiple datasets, the performance improvement of the three-gene model compared to single-gene, dual-gene, and traditional CDKN2A+MKI67 biomarker combinations was evaluated, indicating that the three-gene combination has a synergistic gain effect.

[0102] This combination was determined through a hierarchical comparison of hundreds of candidate genes, hundreds of gene combinations, and thousands of machine learning models, using cross-dataset performance constraints and low-dimensional model comparisons. Its performance improvement stems from the joint contribution of expression signals with different directions and related structures in the ridge regression model, rather than simply selecting the gene with the highest single-gene AUC, or performing conventional parallel combinations of known biomarkers.

[0103] like Figure 1 The image shows a comparison of the ROC curves of different models in the training set:

[0104] 1) Single gene model: The AUC of the SYNGR1 single gene model was 0.894, the sensitivity was 78.2%, and the specificity was 90.8%; the AUC of the TOPBP1 single gene model was 0.913, the sensitivity was 80.8%, and the specificity was 96.9%.

[0105] 2) Traditional biomarker combination: The CDKN2A+MKI67 combined model has an AUC of 0.884, a sensitivity of 79.5%, and a specificity of 100%.

[0106] 3) Dual-gene model: The AUC of the MIF+SYNGR1 dual-gene model was 0.896, the sensitivity was 75.6%, and the specificity was 92.3%; the AUC of the MIF+TOPBP1 dual-gene model was 0.905, the sensitivity was 78.2%, and the specificity was 96.9%.

[0107] 4) The three-gene model of this invention: The AUC of the MIF-SYNGR1-TOPBP1 three-gene model is 0.923, the sensitivity is 91.0%, and the specificity is 72.3%.

[0108] The ROC curve analysis results show that the MIF-SYNGR1-TOPBP1 trigene model of this invention achieved the highest AUC value and sensitivity among all comparative models, indicating that it can more effectively identify high-risk lesions (HSIL and cervical cancer) and reduce the risk of missed diagnoses. Although the specificity of this model at the optimal classification threshold is lower than that of some single-gene models and the traditional CDKN2A+MKI67 biomarker combination, its overall classification performance (AUC) is the best. This indicates that the combined expression of MIF, SYNGR1, and TOPBP1 has a better comprehensive auxiliary judgment ability than single-gene models, dual-gene models, and traditional biomarker combinations, and can provide a more reliable molecular basis for the early screening and auxiliary judgment of high-grade cervical lesions and cervical cancer.

[0109] like Figure 2 As shown, the ROC curves of different models in the TCGA independent external validation set are compared:

[0110] 1) Single gene model: The AUC of the SYNGR1 single gene model was 0.909, the sensitivity was 1.0%, and the specificity was 100%; the AUC of the TOPBP1 single gene model was 0.976, the sensitivity was 100%, and the specificity was 0%.

[0111] 2) Traditional biomarker combination: The CDKN2A+MKI67 combined model has an AUC of 1.000, a sensitivity of 100%, and a specificity of 66.7%.

[0112] 3) Dual-gene model: The AUC of the MIF+SYNGR1 dual-gene model was 0.909, the sensitivity was 0%, and the specificity was 100%; the AUC of the MIF+TOPBP1 dual-gene model was 0.978, the sensitivity was 100%, and the specificity was 0%.

[0113] 4) The three-gene model of this invention: The AUC of the MIF-SYNGR1-TOPBP1 three-gene model is 0.991, the sensitivity is 89.1%, and the specificity is 100%.

[0114] like Figure 3 As shown, the ROC curves of different models in the GSE113942 independent external validation set are compared:

[0115] 1) Single gene model: The AUC of the SYNGR1 single gene model was 0.837, the sensitivity was 0%, and the specificity was 100%; the AUC of the TOPBP1 single gene model was 0.857, the sensitivity was 100%, and the specificity was 0%.

[0116] 2) Traditional biomarker combination: The AUC of the CDKN2A+MKI67 combined model was 1.000, the sensitivity was 100%, and the specificity was 0%.

[0117] 3) Dual-gene model: The AUC of the MIF+SYNGR1 dual-gene model was 0.837, with a sensitivity of 0% and a specificity of 100%; the AUC of the MIF+TOPBP1 dual-gene model was 0.857, with a sensitivity of 100% and a specificity of 0%.

[0118] 4) The three-gene model of this invention: The AUC of the MIF-SYNGR1-TOPBP1 three-gene model is 0.837, the sensitivity is 85.7%, and the specificity is 85.7%.

[0119] The ROC curve analysis results show that, in the GSE113942 independent external validation set, the MIF-SYNGR1-TOPBP1 three-gene model of this invention achieved an AUC value of 0.837. Although its AUC is comparable to some comparative models, the three-gene model of this invention simultaneously achieved a sensitivity of 85.7% and a specificity of 85.7%, achieving a good balance between the detection rate of high-risk lesions and the accurate identification of low-risk lesions. In contrast, some single-gene models, dual-gene models, and traditional CDKN2A+MKI67 biomarker combinations, while having higher AUC values, exhibited lower sensitivity or specificity, resulting in unbalanced classification performance. Therefore, the three-gene co-expression mode of this invention demonstrated good comprehensive auxiliary judgment ability and cross-platform stability in this independent external validation set, providing reliable molecular evidence for the auxiliary judgment of high-grade cervical lesions and cervical cancer.

[0120] like Figure 4 As shown, the ROC curves of different models in the GSE9750 independent external validation set are compared:

[0121] 1) Single gene model: The AUC of the SYNGR1 single gene model was 0.873, the sensitivity was 0%, and the specificity was 100%; the AUC of the TOPBP1 single gene model was 0.965, the sensitivity was 100%, and the specificity was 0%.

[0122] 2) Traditional biomarker combination: The AUC of the CDKN2A+MKI67 combined model was 0.974, the sensitivity was 100%, and the specificity was 0%.

[0123] 3) Dual-gene model: The AUC of the MIF+SYNGR1 dual-gene model was 0.864, with a sensitivity of 0% and a specificity of 100%; the AUC of the MIF+TOPBP1 dual-gene model was 0.963, with a sensitivity of 100% and a specificity of 0%.

[0124] 4) The three-gene model of this invention: The AUC of the MIF-SYNGR1-TOPBP1 three-gene model is 0.955, the sensitivity is 96.3%, and the specificity is 75.0%.

[0125] The comprehensive ROC curve analysis results show that, in the GSE9750 independent external validation set, the MIF-SYNGR1-TOPBP1 tri-gene model of this invention achieved an AUC value of 0.955, demonstrating excellent overall classification performance. Compared with single-gene models, dual-gene models, and the traditional CDKN2A+MKI67 biomarker combination, the tri-gene model of this invention maintains both high sensitivity (96.3%) and good specificity (75.0%), effectively balancing the detection rate of high-risk lesions with the accurate identification of low-risk lesions. In contrast, while some comparative models have high AUC values, their sensitivity or specificity is low, resulting in unbalanced classification performance. Therefore, the tri-gene co-expression mode of this invention demonstrates good comprehensive auxiliary judgment ability and cross-platform stability in this independent external validation set, providing reliable molecular evidence for the auxiliary judgment of high-grade cervical lesions and cervical cancer.

[0126] like Figure 5 As shown, the ROC curves of different models in the GSE27678 independent external validation set are compared:

[0127] 1) Single gene model: The AUC of the SYNGR1 single gene model was 0.834, the sensitivity was 0%, and the specificity was 100%; the AUC of the TOPBP1 single gene model was 0.729, the sensitivity was 100%, and the specificity was 0%.

[0128] 2) Traditional biomarker combination: The AUC of the CDKN2A+MKI67 combined model was 0.917, the sensitivity was 100%, and the specificity was 0%.

[0129] 3) Dual-gene model: The AUC of the MIF+SYNGR1 dual-gene model was 0.836, with a sensitivity of 0% and a specificity of 100%; the AUC of the MIF+TOPBP1 dual-gene model was 0.757, with a sensitivity of 100% and a specificity of 0%.

[0130] 4) The three-gene model of this invention: The AUC of the MIF-SYNGR1-TOPBP1 three-gene model is 0.896, the sensitivity is 89.8%, and the specificity is 69.2%.

[0131] The ROC curve analysis results show that, in the GSE27678 independent external validation set, the MIF-SYNGR1-TOPBP1 three-gene model of this invention achieved an AUC of 0.896. Although slightly lower than the traditional CDKN2A+MKI67 combined model (AUC=0.917), it maintained both high sensitivity (89.8%) and certain specificity (69.2%), demonstrating good overall judgment performance. In contrast, while some single-gene and dual-gene models showed 100% sensitivity or 100% specificity at the optimal classification threshold, another indicator was close to 0%, indicating a significant bias in their classification ability, making it difficult to balance the detection rate and misdiagnosis rate of high-risk lesions. The three-gene model of this invention achieves a good balance between sensitivity and specificity, possessing a more stable and reliable comprehensive auxiliary judgment ability, further validating the good generalization ability and clinical application potential of the MIF, SYNGR1, and TOPBP1 co-expression pattern in different external cohorts.

[0132] like Figure 6 As shown, the ROC curves of different models in the GSE63514 independent external validation set are compared:

[0133] 1) Single gene model: The AUC of the SYNGR1 single gene model was 0.664, the sensitivity was 0%, and the specificity was 100%; the AUC of the TOPBP1 single gene model was 0.850, the sensitivity was 100%, and the specificity was 0%.

[0134] 2) Traditional biomarker combination: The AUC of the CDKN2A+MKI67 combined model was 0.762, the sensitivity was 100%, and the specificity was 0%.

[0135] 3) Dual-gene model: The AUC of the MIF+SYNGR1 dual-gene model was 0.674, with a sensitivity of 0% and a specificity of 100%; the AUC of the MIF+TOPBP1 dual-gene model was 0.859, with a sensitivity of 100% and a specificity of 0%.

[0136] 4) The three-gene model of this invention: The AUC of the MIF-SYNGR1-TOPBP1 three-gene model is 0.802, the sensitivity is 77.9%, and the specificity is 63.3%.

[0137] The ROC curve analysis results show that, in the GSE63514 independent external validation set, the MIF-SYNGR1-TOPBP1 three-gene model of this invention achieved an AUC of 0.802, demonstrating excellent overall classification performance. In comparison, although the AUC (0.802) of the MIF-SYNGR1-TOPBP1 three-gene model of this invention is slightly lower than that of the TOPBP1 single-gene model and the MIF+TOPBP1 dual-gene model, it maintains high sensitivity (77.9%) and specificity (63.3%), exhibiting a more balanced and stable comprehensive discriminative ability. These results indicate that, with the introduction of MIF and SYNGR1, the three-gene combined model can effectively improve the classification bias caused by relying solely on TOPBP1, and has better generalization performance and potential clinical application value in the external independent validation cohort.

[0138] like Figure 7 As shown, the ROC curves of the three-gene model of this invention in the independent clinical external validation set are displayed: the AUC of the MIF-SYNGR1-TOPBP1 three-gene model is 0.922, the sensitivity is 86%, and the specificity is 80.6%.

[0139] In an independent external validation set, the MIF-SYNGR1-TOPBP1 three-gene model of this invention achieved an AUC of 0.922, while maintaining high sensitivity (86%) and specificity (80.6%), demonstrating excellent and stable comprehensive discriminative ability. These results indicate that the model has good generalization performance and clinical application value. Cervical cytology sampling is simple and minimally invasive, suitable for outpatient screening and repeat testing, and has better clinical accessibility compared to tissue sampling. This model only detects three genes—MIF, SYNGR1, and TOPBP1—and can be performed using a conventional RT-qPCR platform, featuring fewer detection indicators, a simple experimental procedure, a short detection cycle, and low conversion costs. The above results demonstrate that this model not only has good risk discrimination ability but also possesses advantages such as easy sample acquisition, widespread availability of testing platforms, a simplified model structure, and convenient clinical deployment, providing a potential molecular detection tool for outpatient auxiliary screening and risk stratification of high-grade cervical lesions and cervical cancer.

[0140] The present invention also provides an auxiliary diagnostic system for high-grade cervical lesions and cervical cancer based on the aforementioned biomarker combination. Preferably, the system includes:

[0141] The data acquisition module is used to acquire data on MIF, SYNGR1, and TOPBP1 in the sample to be tested;

[0142] The data processing module is used to standardize the expressed data;

[0143] The risk assessment module is used to calculate the category to which a sample belongs based on a preset classification model;

[0144] The result output module is used to output auxiliary judgment results.

[0145] To enable portable deployment and cross-platform application of the above-mentioned system functions, the present invention also provides a computer-readable storage medium.

[0146] The computer-readable storage medium stores a computer program, which, when executed by a processor, performs the following steps:

[0147] S1. Obtain the expression data of MIF, SYNGR1 and TOPBP1 in the sample to be tested;

[0148] S2. Standardize the expressed data;

[0149] S3. Input the standardized expression data into the pre-built classification model;

[0150] S4. Determine whether the sample to be tested belongs to the low-risk lesion group or the high-risk lesion group based on the model output results.

[0151] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0152] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A combination of biomarkers for the auxiliary diagnosis of cervical lesions, characterized in that, include: MIF, SYNGR1, and TOPBP1.

2. The biomarker combination for auxiliary diagnosis of cervical lesions according to claim 1, characterized in that: The biomarker combination is used to help identify patients with high-grade squamous intraepithelial lesions and cervical cancer.

3. The biomarker combination for auxiliary diagnosis of cervical lesions according to claim 1, characterized in that: The biomarker combination is used to distinguish between low-risk and high-risk lesion groups, wherein the low-risk lesions include normal cervical tissue and low-grade squamous intraepithelial lesions, and the high-risk lesions include high-grade squamous intraepithelial lesions and cervical cancer tissue.

4. A detection kit prepared based on the combination of biomarkers according to any one of claims 1 to 3, characterized in that, include: Primers and / or probes for detecting MIF expression levels; Primers and / or probes for detecting SYNGR1 expression levels; Primers and / or probes for detecting TOPBP1 expression levels; Internal reference gene detection components: And the amplification reaction system.

5. An application of the combination of biomarkers according to any one of claims 1 to 3 in the preparation of a product for assisting in the diagnosis of cervical lesions, wherein the product is a test kit, a gene chip, or a sequencing library.

6. A method for assisting in the diagnosis of cervical lesions based on a combination of biomarkers according to any one of claims 1 to 3, characterized in that, Includes the following steps: Obtain expression data of MIF, SYNGR1, and TOPBP1 in the sample to be tested; The expressed data are standardized. The standardized representation data is input into a pre-built classification model; The model output results determine whether the sample belongs to the low-risk lesion group or the high-risk lesion group.

7. The method for auxiliary diagnosis of cervical lesions according to claim 6, characterized in that: The method for constructing the classification model includes the following steps: Single-cell transcriptome data of training samples were obtained, and pseudo-bulkRNA-seq expression matrix was constructed using the AggregateExpression function; Logarithmic transformation was performed on the pseudo-bulk RNA-seq expression matrix, and then centered and normalized based on the training set. Using the expression values ​​of MIF, SYNGR1, and TOPBP1 as input variables and sample grouping as the outcome variable, a binary logistic regression model, i.e., a classification model, is constructed using machine learning algorithms.

8. The method for auxiliary diagnosis of cervical lesions according to claim 7, characterized in that: The machine learning algorithms include one or more of glmnet-ridge, ridge logistic regression, lasso logistic regression, elastic-net logistic regression, random forest, support vector machine, or xgboost.

9. A system for assisting in the diagnosis of cervical lesions, characterized in that, include: The data acquisition module is used to acquire data on MIF, SYNGR1, and TOPBP1 in the sample to be tested; The data processing module is used to standardize the expressed data; The risk assessment module is used to calculate the category to which a sample belongs based on a preset classification model; The result output module is used to output auxiliary judgment results.

10. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the auxiliary judgment method of claim 6.

Images