Marker combination for grading noninvasive risk degree of neuroblastoma, prediction model and prediction method and application thereof

Abstract

The invention belongs to the technical field of bioinformatics and medical detection, and particularly relates to a marker combination for neuroblastoma (NB) noninvasive risk level grading, a prediction model, a prediction method and application thereof. The marker combination is used for determining the sex, determining whether the month age is greater than 18 months, determining whether plasma MYCN is amplified, determining whether tumors are metastatic, and determining the content of neuron-specific enolase and lactic dehydrogenase; a machine learning algorithm is used for constructing an NB noninvasive risk degree grading prediction model, the comprehensive performance of the random forest model is optimal, the area value under a subject working characteristic curve reaches 0.956, the sensitivity is 92.9%, the specificity is 82.1%, the accuracy rate is 87.5%, the Kappa value is 0.75, the F1 score is 0.881, and NB middle and low risk patients and NB high risk patients can be effectively distinguished; the NB non-invasive risk level grading prediction model constructed by the invention can quickly, accurately and non-invasively perform NB risk level grading, and has a relatively good clinical application value.

Description

Technical Field

[0001] This invention belongs to the field of bioinformatics and medical testing technology, and specifically relates to a combination of markers and a predictive model for non-invasive risk grading of neuroblastoma. Background Technology

[0002] Neuroblastoma (NB) is one of the most common solid tumors in children, exhibiting highly heterogeneous clinical phenotypes and significant differences in prognosis among patients. Approximately half of all patients are diagnosed with high-risk NB, and even with intensive multimodal therapy, the 5-year overall survival rate is only 48%. Risk stratification serves as a crucial bridge connecting clinical diagnosis and precision medicine. Its core value lies in developing individualized treatment plans for patients with different risk stratifications, reducing the risk of overtreatment or undertreatment, and is essential for improving patient outcomes.

[0003] Currently, the Childhood Oncology Group (COG) in the United States is based on MYCN Key indicators such as copy number, DNA ploidy, International Nephrotic Biology Scheme (INSS), and International Nephrotic Biology Pathological Classification (INPC) classify nephrotic bile into low-risk, intermediate-risk, and high-risk categories. However, some of these classification indicators, such as MYCN copy number, DNA ploidy, and INPC, require surgical resection or biopsy to obtain nephrotic bile tissue samples for testing. This invasive sampling-dependent testing method has significant limitations in clinical application. Clinically, most high-risk nephrotic bile patients are already in the late stage of the disease when they are diagnosed, with large tumors often accompanied by widespread metastasis. It is impossible to safely obtain sufficient tumor tissue through surgery or biopsy, resulting in the inability to complete risk classification in a timely manner, which in turn delays the formulation of individualized treatment plans and seriously affects the patient's treatment outcome and survival prognosis.

[0004] Patent 202311312440.6 discloses a prognostic model for neonatal tumor (NB), which integrates differentially amplified MYCN genes and cellular senescence genes to construct a multi-gene combination model. This model is more accurate in predicting NB prognosis than a single high-risk gene and can assess the tumor microenvironment of immunosuppression and senescence. Verification has shown that the prognostic model constructed in this patent exhibits excellent performance in predicting 3-year and 5-year overall survival (OS) and event-free survival (EFS). However, this technology only focuses on prognostic prediction and tumor microenvironment assessment, and does not optimize the design for non-invasive NB risk stratification. It fails to overcome the dependence of traditional stratification systems on invasive tissue samples. Therefore, establishing a new NB risk stratification strategy based on non-invasive detection technology to achieve timely and accurate risk stratification for children with advanced disease and those for whom tissue samples cannot be obtained, and to provide support for the timely formulation of individualized treatment plans, has become a key issue that urgently needs to be addressed in clinical practice. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a combination of markers, a prediction model, a prediction method, and applications for non-invasive risk grading of neuroblastoma.

[0006] The technical solution of this invention is implemented as follows:

[0007] This invention systematically screens clinical NB (not invasive) detection indicators, confirming that six indicators are promising for NB risk stratification. Furthermore, based on machine learning algorithms, it identifies a combination of six non-invasive indicators using a random forest algorithm as the most effective method for NB risk stratification.

[0008] Based on this, the present invention provides a combination of markers for non-invasive risk grading of neuroblastoma, the combination of markers including sex, age (greater than 18 months), and plasma... MYCN Whether the tumor has amplified, whether it has metastasized, the levels of neuron-specific enolase and lactate dehydrogenase. To achieve non-invasive risk grading for neonatal invasive surgery (NB), this invention... MYCN The amplified state uses free plasma MYCN Gene amplification status.

[0009] Secondly, the present invention also provides a method for constructing a non-invasive risk grading prediction model for neuroblastoma, comprising the following steps:

[0010] (1) Collect clinical data of children who are diagnosed with neuroblastoma for the first time, construct a feature dataset containing only the above-mentioned combination of markers, and form a sample set;

[0011] (2) Divide the sample set into a training set and a test set; use the training set to train multiple basic machine learning classification models respectively;

[0012] (3) The performance of each model was evaluated using the test set, the ROC curve of each model was plotted, the core performance indicators were extracted to evaluate the overall performance of each model, and the model with the best performance was selected as the non-invasive risk grading prediction model for neuroblastoma.

[0013] The screening criteria for clinical data are as follows:

[0014] ① A confirmed diagnosis of NB, excluding cases with serious complications (severe pneumonia, sepsis, or other highly inflammatory states); ② The first confirmed case; ③ FISH assessment. MYCN ④ Amplification status; ⑤ Plasma sample available; ⑥ Written informed consent signed by the child's guardian at the time of sample collection.

[0015] Preferably, the aforementioned basic machine learning classification models include logistic regression models, linear discriminant analysis models, elastic network regression models, support vector machines, K-nearest neighbors models, random forest models, multilayer perceptron models, neural network models, Naive Bayes models, flexible discriminant analysis models, and classification regression tree models.

[0016] Preferably, the aforementioned core performance indicators include AUC value, sensitivity, specificity, accuracy, Kappa value, and F1 score.

[0017] The training set achieved a sensitivity of 87.5%, a specificity of 84.0%, and an overall accuracy of 85.7%, with positive predictive value (PPV) and negative predictive value (NPV) of 84.5% and 87.0%, respectively.

[0018] The validation set had a sensitivity of 91.7% and a specificity of 83.3%, while the overall accuracy was 87.5%, with positive predictive value (PPV) and negative predictive value (NPV) of 84.6% and 90.9%, respectively.

[0019] Preferably, the best-performing model is the random forest model. Its AUC is 0.956, sensitivity is 92.9%, specificity is 82.1%, accuracy is 87.5%, Kappa value is 0.75, and F1 score is 0.881.

[0020] Thirdly, the above-mentioned construction method constructs a non-invasive risk grading prediction model for neuroblastoma.

[0021] Fourthly, the above-mentioned non-invasive risk grading prediction model for neuroblastoma is applied to the prediction of non-invasive risk grading for neuroblastoma.

[0022] Fifthly, a predictive method for non-invasive risk grading of neuroblastoma, comprising the following steps:

[0023] 1) Collect feature data of marker combinations in the sample to be tested to form a joint non-invasive feature set;

[0024] 2) Input the joint non-invasive feature set into the random forest model, and determine the risk level of the sample to be tested based on the optimal cut-off value of the random forest model.

[0025] Preferably, the optimal cut-off value mentioned above is based on the Youden index as the evaluation standard, and the value corresponding to the largest Youden index is selected as the optimal cut-off; where the Youden index = sensitivity + specificity - 1. Here, the sensitivity is defined as >0.7. At this point, when the Youden index is at its maximum, both sensitivity and specificity are relatively large, and this point on the ROC curve is the most suitable point that balances the best sensitivity and best specificity.

[0026] Furthermore, if the risk score of the test sample output by the random forest model is higher than the optimal cut-off value, it is identified as a high-risk sample; if the risk score of the test sample output by the random forest model is lower than the optimal cut-off value, it is identified as a medium- or low-risk sample.

[0027] The present invention has the following beneficial effects:

[0028] This invention provides a combination of markers for non-invasive risk grading of neuroblastoma (NB) to construct an accurate and clinically valuable risk grading strategy, including sex, age (greater than 18 months), and plasma risk markers. MYCN Whether the tumor has expanded, whether it has metastasized, the levels of neuron-specific enolase and lactate dehydrogenase, including plasma levels. MYCN The markers were confirmed to be consistent with the MYCN amplification status in the tissue. Based on the above marker combination, this application constructed a non-invasive risk grading prediction model for neonatal nephropathy (NB) using machine learning algorithms. The random forest (RF) model showed the best overall performance, with an AUC of 0.956. It demonstrated optimal performance in core indicators such as sensitivity (92.9%), specificity (82.1%), and accuracy (87.5%). The optimal cut-off value based on the random forest model effectively distinguished between low- and high-risk NB patients. Based on confusion matrix analysis, the training set model had a sensitivity of 87.5%, a specificity of 84.0%, and an overall accuracy of 85.7%, with positive predictive value (PPV) and negative predictive value (NPV) of 84.5% and 87.0%, respectively. The validation set model had a sensitivity of 91.7%, a specificity of 83.3%, and an overall accuracy of 87.5%, with PPV and NPV of 84.6% and 90.9%, respectively. Therefore, the non-invasive risk grading prediction model for NB constructed in this application has good clinical application value. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 ROC curves for 9 candidate indicators used in NB risk stratification; where Figure A is the ROC curve for sex, B is the ROC curve for whether the baby is older than 18 months, and C is... MYCNROC curves for gene amplification status, D for tumor metastasis, E for S100, F for IL-6, G for NSE, H for LDH, and I for PCT.

[0031] Figure 2 The ROC curves of the non-invasive prediction model combining 6 clinical indicators were compared after analysis by 11 different machine learning algorithms. The horizontal axis represents the false positive rate, the vertical axis represents the true positive rate, AUC is the area under the receiver operating characteristic curve, and 95% CI is the 95% confidence interval of the AUC value.

[0032] Figure 3 ROC curves combining six clinical indicators were used to differentiate between low- and intermediate-risk NB and high-risk NB.

[0033] Figure 4 The confusion matrix analysis was used to show the combined diagnostic efficacy of six clinical indicators; Figure A shows the diagnostic efficacy of the training set, and Figure B shows the diagnostic efficacy of the validation set.

[0034] Figure 5 This invention relates to a non-invasive NB risk grading prediction platform; Figure A shows the platform's homepage, and Figure B shows an example of the prediction results page.

[0035] Figure 6 This is a demonstration of the operation of the NB non-invasive risk grading prediction platform website; where Figure A is example a and Figure B is example b. Detailed Implementation

[0036] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0037] Unless otherwise specified, the experimental methods used in the following experimental examples are conventional methods; the materials and reagents used are commercially available unless otherwise specified.

[0038] Example 1: Predictive Value of Clinical Indicators in NB Risk Staging

[0039] 1. Screening of clinical indicators

[0040] This application collected clinical data from 160 children with neonatal nephropathy (NB) in the Hepatobiliary and Oncology Surgery Department of Henan Provincial Children's Hospital and Zhengzhou Children's Hospital. The inclusion criteria are as follows:

[0041] (1) Cases with a confirmed diagnosis of NB, excluding those with serious complications (severe pneumonia, sepsis, and other highly inflammatory states);

[0042] (2) First confirmed case;

[0043] (3) FISH method evaluation MYCN Amplification state;

[0044] (4) A plasma sample is available;

[0045] (5) The child’s guardian shall sign a written informed consent form when the sample is collected.

[0046] 2. Validation of Independent Predictive Performance of Single Indicators

[0047] To construct a precise and clinically applicable risk stratification strategy for neonatal nephropathy (NB), our research team, based on actual clinical needs, extensively explored and initially identified nine candidate indicators from four dimensions: serological testing, molecular diagnosis, clinical information, and imaging assessment. These indicators include gender, age, and whether the tumor has metastasized as defined by imaging (hereinafter referred to as whether the tumor has metastasized). MYCN Gene amplification status, including S100 calcium-binding protein (S100), interleukin-6 (IL-6), procalcitonin (PCT), neuron-specific enolase (NSE), and lactate dehydrogenase (LDH). To address the issue of missing sample data for some serological candidate indicators (S100, IL-6, PCT), this study employed various mainstream missing value handling strategies, including mean imputation, multiple imputation, and K-nearest neighbor (KNN) imputation, to maximize data integrity. Simultaneously, statistical methods such as univariate logistic regression analysis, receiver operating characteristic (ROC) curve and area under the curve (AUC) calculation, and intergroup significance tests were used to validate the hierarchical efficacy of each indicator from multiple dimensions.

[0048] To ensure the scientific rigor, robustness, and clinical applicability of the grading system, multiple rounds of statistical screening and clinical relevance verification were conducted. Ultimately, the three indicators (S100, IL-6, PCT) that did not meet expectations were eliminated, resulting in the selection of six core indicators with significant discriminative power (gender, age, tumor metastasis, etc.). MYCN Gene amplification status, NSE, LDH), as basic features for subsequent modeling ( Figure 1 For the six non-invasive indicators selected, receiver operating characteristic (ROC) curves were plotted using the pROC package in R language. The area under the curve (AUC) of each indicator in predicting the risk of non-invasive brain injury (NIB) (low to intermediate / high risk) was calculated to evaluate the independent predictive value of individual indicators and provide a reference for subsequent multi-indicator joint modeling.

[0049] Six indicators that were both non-invasive and statistically significant (P < 0.05) were successfully screened: sex, age (greater than 18 months), and plasma concentration. MYCN Gene amplification status, tumor metastasis, NSE, and LDH were all considered. Single-indicator prediction results showed that all six non-invasive indicators demonstrated some predictive ability for NB risk. Among them, the single-factor model corresponding to tumor metastasis showed relatively better predictive efficacy, with an area under the ROC curve (AUC) of 0.848, which can, to some extent, distinguish between high-risk and low-risk NB children. Figure 1 However, the predictive performance of a single indicator is limited and does not meet the requirements for high-precision clinical applications.

[0050] Example 2: Plasma MYCN Copy number quantification method

[0051] To achieve non-invasive NB risk stratification, this embodiment evaluated the plasma free MYCN copy number of NB infants and compared it with the clinically measured FISH-based method. MYCN Copy number consistency.

[0052] (1) Extraction of cell-free DNA from plasma: Cell-free DNA was extracted from plasma using the blood / cell / tissue genomic DNA extraction kit (DP304) from Tiangen Biotech (Beijing) Co., Ltd.

[0053] (2) RT-PCR detection of plasma MYCN Copy number. The detection method was based on the paper published by the applicant team (CancerReports 2023, 6(2): e1688.), and the RT-PCR detection results were compared with the clinical FISH results.

[0054] Table 1. Representative NB tissues and plasma MYCN Amplification Status Correspondence Result Table

[0055]

[0056] The results are shown in Table 1. In NB patients, plasma and tissue samples showed... MYCN The amplification status is consistent. Therefore, in order to achieve non-invasive risk stratification of NB, this invention... MYCN The amplified state uses free plasma MYCN Gene amplification status.

[0057] Example 3: Machine Learning Model Construction and Training

[0058] Six non-invasive indicators were combined to construct a predictive feature set containing only clinical non-invasive indicators. Eleven commonly used machine learning classification models were selected, including Logistic Regression (LR), Linear Discriminant Analysis (LDA), Elastic Net (Enet), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), Multilayer Perceptron (MLP), Neural Network (NN), Naive Bayes (NB), Flexible Discriminant Analysis (FDA), and Classification and Regression Tree (CART). These six selected non-invasive feature variables were combined to form a unified input feature set. Based on this feature set, eleven machine learning classifiers were constructed, resulting in eleven classification prediction schemes based on a single non-invasive feature set and different algorithm models. ROC curves for the eleven algorithms were plotted. Figure 2 The core performance metrics of each model (AUC, sensitivity, specificity, accuracy, Kappa value, F1 score, etc.) are extracted and compared comprehensively; based on the overall performance of the metrics, the optimal prediction model is selected.

[0059] Confusion matrix construction and model performance evaluation: For the optimal model (RF model) selected through screening, the confusion matrix is ​​calculated to quantify the true positive, true negative, false positive, and false negative results of the model in the low-to-medium risk / high-risk classification, and to further verify the classification reliability and clinical application value of the model.

[0060] Table 2. Results of the non-invasive predictive model combining 6 clinical indicators analyzed using 11 different machine learning algorithms.

[0061]

[0062] The Best Cut-off value is defined as the distance from each point on the ROC curve to the top-left corner (0, 1). The point with the smallest distance is the optimal cut-off value. This point maximizes both sensitivity and specificity. The Youden index (Youden index = sensitivity + specificity - 1) is used as the evaluation criterion, and the value corresponding to the largest Youden index is selected as the optimal cut-off. Here, sensitivity > 0.7 is defined. When the Youden index is at its maximum, both sensitivity and specificity are relatively high, and this point on the ROC curve represents the most suitable point that balances optimal sensitivity and specificity.

[0063] To further improve the sensitivity and specificity of noninvasive nephropathy (NB) risk stratification, this study constructed a NB noninvasive risk stratification prediction model using machine learning algorithms based on the above six indicators. Among them, the random forest (RF) model showed the best overall performance, with an area under the ROC curve (AUC) of 0.956, significantly higher than other algorithm models. It also demonstrated the best performance in core indicators such as sensitivity (92.9%), specificity (82.1%), and accuracy (87.5%), effectively distinguishing between low- and high-risk NB patients (Table 2). Figure 3 Based on confusion matrix analysis, the training set model achieved a sensitivity of 87.5%, a specificity of 84.0%, and an overall accuracy of 85.7%, with positive predictive value (PPV) and negative predictive value (NPV) of 84.5% and 87.0%, respectively. The validation set model achieved a sensitivity of 91.7%, a specificity of 83.3%, and an overall accuracy of 87.5%, with positive predictive value (PPV) and negative predictive value (NPV) of 84.6% and 90.9%, respectively. Figure 4 Therefore, the non-invasive risk grading prediction model for NB constructed in this study has good clinical application value.

[0064] Application example: Use of the NB non-invasive risk grading prediction model

[0065] To facilitate the use of this model, we have constructed a "Non-invasive Risk Grading Prediction Platform for Neuroblastoma," the specific usage method of which is as follows:

[0066] (1) Open the website, the URL is zhoaliang123.shinyapps.io / shiny_app / .

[0067] (2) Website homepage

[0068] The specific website homepage is as follows Figure 5 As shown in Figure A.

[0069] (3) Website operation method

[0070] Enter the data into the website homepage in sequence ( Figure 5 (A) Enter the child's real information in the red box section, click Start Calculation, and the reference values ​​for the 6 constructed indicators are shown in Table 3.

[0071] Table 3 Reference values ​​for 6 construction indicators

[0072]

[0073] (4) Interpretation of website results:

[0074] Clicking "Start Calculation" will generate a corresponding prediction report, as shown below:

[0075] Example of prediction results is as follows ( Figure 5 B):

[0076] Predicted HR probability: 0.074 (This suggests a 7.4% probability of high risk in the neuroblastoma risk classification of this child).

[0077] Best cutoff: 0.369

[0078] Risk classification: Low / Intermediate Risk (This indicates that the child's neuroblastoma risk level is low to intermediate).

[0079] Interpretation: The patient is predicted to have low or intermediate risk of neuroblastoma.

[0080] (5) Website operation demonstration

[0081] Example a: Taking actual low-risk neuroblastoma patients as an example:

[0082] 1) Data Input

[0083] The data input is as follows:

[0084] ① Gender: Male 1;

[0085] ② Is the age greater than 18 months: 1 if yes;

[0086] ③Plasma MYCN Gene amplification status: No amplification: 0;

[0087] ④ Whether the tumor has metastasized: 0 for non-metastasis;

[0088] ⑤ Neuron-specific enolase: 14.6 ng / mL;

[0089] ⑥ Lactate dehydrogenase: 261 U / L;

[0090] 2) Click Calculate

[0091] Click "Start Calculation"

[0092] 3) Generate report

[0093] After clicking "Start Calculation", a "Prediction Results" report will be automatically generated, and the results are as follows: Figure 6 As shown in A:

[0094] The results showed that the predicted risk for this child with neuroblastoma was low to intermediate.

[0095] Example b: Taking actual high-risk neuroblastoma patients as an example:

[0096] 1) Data Input

[0097] The data input is as follows:

[0098] ① Gender: Male 1;

[0099] ②Is the age greater than 18 months: Yes 1;

[0100] ③Plasma MYCN Gene amplification: Amplification 1;

[0101] ④ Whether the tumor has metastasized: Metastasis 1;

[0102] ⑤ Neuron-specific enolase: 370 ng / mL;

[0103] ⑥ Lactate dehydrogenase: 389 U / L;

[0104] 2) Click Calculate

[0105] Click "Start Calculation"

[0106] 3) Generate report

[0107] After clicking "Start Calculation", a "Prediction Results" report will be automatically generated, and the results are as follows: Figure 6 As shown in B:

[0108] The results indicate that the predicted risk for this child with neuroblastoma is high. The patient is expected to be at higher risk and may require closer monitoring.

[0109] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A combination of markers for non-invasive risk grading of neuroblastoma, characterized in that, The marker combination includes sex, age (greater than 18 months), and plasma... MYCN Whether it amplifies, whether the tumor has metastasized, the content of neuron-specific enolase and lactate dehydrogenase.

2. A method for constructing a non-invasive risk grading prediction model for neuroblastoma, characterized in that, Includes the following steps: (1) Collect clinical data of children who are diagnosed with neuroblastoma for the first time, construct a feature dataset containing only the combination of markers described in claim 1, and form a sample set; (2) Divide the sample set into a training set and a test set; use the training set to train multiple basic machine learning classification models respectively; (3) The performance of each model was evaluated using the test set, the ROC curve of each model was plotted, the core performance indicators were extracted to evaluate the overall performance of each model, and the model with the best performance was selected as the non-invasive risk grading prediction model for neuroblastoma.

3. The construction method according to claim 2, characterized in that: The basic machine learning classification models include logistic regression, linear discriminant analysis, elastic network regression, support vector machine, K-nearest neighbor, random forest, multilayer perceptron, neural network, naive Bayes, flexible discriminant analysis, and classification regression tree models.

4. The construction method according to claim 3, characterized in that: The core performance metrics include AUC value, sensitivity, specificity, accuracy, Kappa value, and F1 score.

5. The construction method according to claim 4, characterized in that: The best-performing model is the random forest model.

6. A non-invasive risk grading prediction model for neuroblastoma constructed using the construction method described in any one of claims 1-5.

7. The application of the non-invasive risk grading prediction model for neuroblastoma as described in claim 6 in the prediction of non-invasive risk grading for neuroblastoma.

8. A method for predicting non-invasive risk grading of neuroblastoma using the marker combination described in claim 1, characterized in that, The steps are as follows: 1) Collect feature data of the combination of markers in the sample to be tested to form a joint non-invasive feature set; 2) Input the joint non-invasive feature set into the random forest model, and determine the risk level of the sample to be tested based on the optimal cut-off value of the random forest model.

9. The prediction method according to claim 8, characterized in that: The optimal cut-off value is evaluated using the Youden index as the evaluation criterion, and the value corresponding to the largest Youden index is selected as the optimal cut-off; where the Youden index = sensitivity + specificity - 1.

10. The prediction method according to claim 9, characterized in that: If the risk score of the test sample output by the random forest model is higher than the optimal cut-off value, it is identified as a high-risk sample; if the risk score of the test sample output by the random forest model is lower than the optimal cut-off value, it is identified as a medium- or low-risk sample.

Images