Classification device for central nervous system tumors and its application

Abstract

The present invention provides a classification device for central nervous system tumors and its application, which relates to the field of molecular biology technology. The training samples used by the classification device for central nervous system tumors provided by the present invention are more complete, covering 148 types of CNS tumors, increasing the diversity of tumor methylation, and ensuring the classification accuracy in practical applications; solving the problem that tumor diagnosis results cannot be made from the morphological level, avoiding human subjectivity, and improving diagnostic efficiency; the results of 3-fold cross-validation show that the classifier of the present invention has high accuracy: the accuracy rate can reach 0.988, and the AUC value is 0.971; it can be used for NGS (second-generation sequencing) data.

Description

Technical Field

[0001] The invention relates to the technical field of molecular biology, and in particular to a classification device for central nervous system tumors and application thereof. Background Art

[0002] Central nervous system tumors (CNS) are a highly lethal disease that occurs frequently in children and originates from the brain and spinal cord. CNS tumors can be divided into more than 100 types based on their different tissue morphological characteristics. In the past, traditional detection methods could not make a correct diagnosis of brain tumors, which often led to poor prognosis. Secondly, precision cancer medicine is also inseparable from accurate pathological diagnosis.

[0003] With the improvement of our understanding of biological molecules, the accuracy of tumor diagnosis is getting higher and higher, and treatment methods are becoming more advanced. In 2021, WHO (World Health Organization) issued the fifth edition of the central nervous system tumor classification system, emphasizing the role of molecular diagnosis in the diagnosis of CNS tumors, and identified new tumor types and subtypes based on DNA methylation maps, which greatly improved the cancer precision medicine strategy and provided clinicians with new treatment directions. The application of DNA methylation in CNS tumor detection will enable more subtypes to be discovered, making treatment more precise and benefiting human society.

[0004] The development of molecular biology has brought convenience to pathological detection, but also brought challenges, especially the rapid development of omics mapping technology in recent years. In order to reduce the burden of human pathological detection and ensure the accuracy of diagnosis, the development of automated diagnostic methods is urgent, and corresponding artificial intelligence technology has emerged. The German Cancer Research Network (DKFZ) used random forest and L2 regularized logistic regression algorithms to construct an automated classifier with excellent diagnostic performance for 91 CNS tumor subtypes: the accuracy of the three-fold cross-validation set was 0.9886; the accuracy of the independent sample set was 88%. Similarly, Northwestern University School of Medicine in the United States used the same method to construct an automated CNS tumor classifier that can be used for clinical testing, and its predictive performance was comparable to that of the DKFZ classifier. Obviously, the automation of pathological diagnosis based on molecular biological characteristics is a development trend. However, my country's technological development in the field of CNS tumor diagnosis is still lacking, especially the application of artificial intelligence technology in this regard.

[0005] In view of this, the present invention is proposed. Summary of the invention

[0006] The first objective of the present invention is to provide a central nervous system tumor classification device to solve the above technical problems.

[0007] The second object of the present invention is to provide the use of the above-mentioned classification device in the preparation of products for diagnosing central nervous system tumor types.

[0008] In order to achieve the above objectives, the following technical solutions are adopted:

[0009] In a first aspect, the present invention provides a classification device for central nervous system tumors, the classification device comprising:

[0010] An acquisition module, used to acquire DNA methylation data of a central nervous system tumor sample to be tested, wherein the DNA methylation data includes a three-dimensional matrix consisting of a sample size, a DNA methylation signal intensity, and a DNA unmethylation signal intensity of the sample to be tested;

[0011] A classification module, used for inputting the DNA methylation data into a pre-trained classifier model, classifying the DNA methylation data through the classifier model, and outputting the classification result of the DNA methylation data; wherein the classifier model is trained based on DNA methylation sample data of multiple preset types, and the classification result includes the probability that the DNA methylation data of the sample to be tested belongs to each preset type;

[0012] A determination module is used to determine the type of DNA methylation data of the sample to be tested based on the classification result.

[0013] As a further technical solution, the preset types include: "Astroblastoma, with MN1 changes, MN1-BEND2 fusion type", "Angiocentric glioma, with MYB / MYBL1 changes", "Astrocytoma, IDH mutant, high grade", "Astrocytoma, IDH mutant, low grade", "Anaplastic neuroepithelial tumor with condensed nuclei", "Rhabdomyosarcoma, alveolar subtype", "Atypical teratoid / rhabdoid tumor, MYC subtype", "Atypical teratoid / rhabdoid tumor, SHH subtype", "Atypical teratoid / rhabdoid tumor, TYR subtype", "Cauda equina neuroendocrine tumor or paraganglioma, non-CIMP subtype", "Chordoid glioma, PRKCA mutations,” “Chordoma,” “CNS neurocytoma,” “CNS tumors with BCOR / BCORL1 fusions,” “CNS tumors with BCOR internal tandem duplications,” “CNS neuroblastoma, FOXR2 activation,” “CIC rearranged sarcoma,” “Primary intracranial sarcoma, DICER1 mutation,” “CNS schwannoma, VGLL fusion,” “Chroid plexus carcinoma,” “Ameloblastic craniopharyngioma,” “Papillary craniopharyngioma,” “Choroid plexus papilloma,” “Cribriform neuroepithelial tumor with SMARCB1 alterations,” “Diffuse glioneuronal neoplasm with oligodendroglioma-like features and nuclear clusters (DGONC),” “Diffuse hemispheric glioma, H3 G34 mutant", "desmoplastic infantile ganglioglioma / desmoplastic infantile astrocytoma", "CNS diffuse large B-cell lymphoma", "diffuse leptomeningeal glioneuronal tumor, subtype 1", "diffuse leptomeningeal glioneuronal tumor, subtype 2", "diffuse midline glioma, H3 K27 alteration, EGFR altered subtype", "diffuse midline glioma, H3 K27 alteration, H3 K27 mutation or EZHIP expression subtype", "dysembryoplastic neuroepithelial tumor", "myxopapillary ependymoma", "posterior fossa type A ependymoma", "posterior fossa type B (PFB) ependymoma", "subependymoma, posterior fossa", "spinal cord ependymoma", "spinal cord ependymoma, MYCN amplification", "subependymoma, spinal cord", "supratentorial subependymoma", "supratentorial ependymoma, ZFTA fusion-positive", "supratentorial ependymoma, ZFTA fusion-positive, ZFTA-RELA fusion subtype, subtype A", "supratentorial ependymoma, ZFTA fusion-positive, ZFTA-RELA fusion subtype, subtype B", "supratentorial ependymoma, YAP1 fusion-positive", "rhabdomyosarcoma, embryonal subtype", "CNS embryonal tumor with BRD4:LEUTX fusion", "embryonal tumor with multilayered rosettes without C19MC changes", "embryonal tumor with multilayered rosettes,C19MC alterations", "CNS embryonal tumor with PLAG family expansion", "Extraventricular neuroblastoma", "Ewing sarcoma", "Glioblastoma, IDH wild-type, posterior fossa subtype", "Glioblastoma, IDH wild-type, mesenchymal subtype, subtype B", "Glioblastoma, IDH wild-type, mesenchymal subtype", "Glioblastoma, IDH wild-type, with primitive neuronal component", "Glioblastoma, IDH wild-type, RTK1 subtype", "Glioblastoma, IDH wild-type, RTK2 subtype", "Germ cell tumor, KIT wild-type subtype", "Germ cell tumor, KIT mutant subtype", "Teratoma", "Yolk sac Tumors", "Glioneuronal tumors, subtype A", "High-grade astrocytoma with lentiform features", "Adult-type diffuse high-grade glioma, IDH wild-type, subtype B", "Adult-type diffuse high-grade glioma, IDH wild-type, subtype E", "Adult-type diffuse high-grade glioma, IDH wild-type, subtype F", "Hemangioblastoma", "Infantile hemispheric glioma", "Inflammatory microenvironment", "Intraocular medulloepithelioma", "Langerhans cell histiocytosis", "Diffuse astrocytoma, MYB or MYBL1 alterations", "Cerebellar liponeurocytoma", "Medulloblastoma, non-WNT / non-SHH, Group 3 Subtype, Subtype I", "Medulloblastoma, non-WNT / non-SHH, Group 3 Subtype, Subtype II", "Medulloblastoma, non-WNT / non-SHH, Group 3 Subtype, Subtype III", "Medulloblastoma, non-WNT / non-SHH, Group 3 Subtype, Subtype IV", "Medulloblastoma, non-WNT / non-SHH, Group 4 Subtype, Subtype V", "Medulloblastoma, non-WNT / non-SHH, Group 4 Subtype, Subtype VI", "Medulloblastoma, non-WNT / non-SHH, Group 4 Subtype, Subtype VII", "Medulloblastoma, non-WNT / non-SHH, Group 4 subtypes, subtype VIII", "Medulloblastoma", "Medulloblastoma, SHH activated, subtype 1", "Medulloblastoma, SHH activated, subtype 2", "Medulloblastoma, SHH activated, subtype 3", "Medulloblastoma, SHH activated, subtype 4", "Medulloblastoma, SHH activated, IDH mutation", "Medulloblastoma, WNT activation", "Melanocytic tumor", "Melanoma metastasis", "Malignant melanoma", "Meningioma, benign", "Meningioma, subtype intermediate", "Meningioma, malignant subtype", "Meningioma, clear cell subtype, SMARCE1 alteration", "Atypical malignant peripheral nerve sheath tumor", "Malignant peripheral nerve sheath tumor", "Myxoid glioneuronal tumor, PDGFRA mutation", "Neuroblastoma, MYCN type", "Neuroblastoma, TMM negative", "Neuroblastoma,ALT / TERT TMM-positive", "Neuroepithelial neoplasm, MN1:CXXC5 fusion", "Neuroepithelial neoplasm with PATZ1 fusion", "Neuroepithelial neoplasm, PLAGL1 fusion", "Plexiform neurofibroma", "Oligodendroglioblastoma, IDH mutation and 1p / 19q co-deletion", "Oligosarcoma, IDH mutant", "Osthonecrosis, IDH wild-type", "Pilocytic astrocytoma, hemispheric", "Pilocytic astrocytoma, infratentorial", "Pilocytic astrocytoma, infratentorial, FGFR1 altered subtype", "Pilocytic astrocytoma, midline", "Pineoblastoma, MYC / FOXR 2 activation", "Pineoblastoma, miRNA processing alteration 1 subtype, subtype A", "Pineoblastoma, miRNA processing alteration 1 subtype, subtype B", "Pineoblastoma, miRNA processing alteration 2 subtype", "Diffuse childhood high-grade glioma, H3 and IDH wild-type, subtype A", "Diffuse childhood high-grade glioma, H3 and IDH wild-type, subtype B", "Diffuse childhood high-grade glioma, MYCN subtype", "Diffuse childhood high-grade glioma, RTK subtype", "Papillary glioneuronal tumor, PRKDC fusion", "Pineocytoma", "Pine Pituitary adenoma, RB1 altered subtype (pineal retinoblastoma)", "Pituitary adenoma, ACTH-secreting subtype", "Pituitary adenoma, gonadotropin-secreting subtype", "Pituitary adenoma, prolactin-secreting subtype", "Pituitary adenoma, STH-secreting subtype, dense granular A subtype", "Pituitary adenoma, STH-secreting subtype, dense granular B subtype", "Pituitary adenoma, STH-secreting subtype, sparse granular subtype", "Pituitary adenoma, TSH-secreting subtype", "Pituitary cell tumor, sellar granular cell tumor, spindle oncocytoma", "CNS plasmacytoma", "Pleomorphic low-grade neuroepithelial tumor of the young", "Moderately differentiated pine "Pineal parenchymal tumor, KBTBD4 alteration, subtype A", "Moderately differentiated pineal parenchymal tumor, KBTBD4 alteration, subtype B", "Pineal region papillary tumor, subtype A", "Pineal region papillary tumor, subtype B", "Pleomorphic xanthoastrocytoma", "Retinoblastoma", "Retinoblastoma, MYCN altered subtype", "Rosette-forming glioneuronal tumor", "Rhabdomyosarcoma, MYOD1 mutant", "Neurilemma", "Subependymal giant cell astrocytoma, TSC1 / TSC2 alteration", "Solitary fibrous tumor", and "Undifferentiated sinonasal carcinoma, IDH2 mutation".

[0014] It should be noted that, among the above types, the names of the various subtypes are names well known in the art, and the “ / ” in the names means “and”.

[0015] As a further technical solution, the DNA methylation data includes DNA methylation data of a DNA methylation chip or second-generation sequencing data;

[0016] The DNA methylation chip includes at least one of the GPL13534, GPL33022 and GPL21145 chips.

[0017] As a further technical solution, the classifier model includes: an input layer, an intermediate layer and an output layer;

[0018] The intermediate layer includes a preset number of convolutional layers, a pooling layer between two adjacent convolutional layers, a flattening layer connected to the last convolutional layer, and a preset number of fully connected layers;

[0019] Wherein, the number of neurons configured in the output layer is consistent with the number of the preset types;

[0020] The training method of the classifier model includes:

[0021] Obtaining a pre-constructed training set, wherein the training set includes DNA methylation sample data of each preset type; and the DNA methylation sample data carries a label corresponding to the preset type;

[0022] The preset type of DNA methylation sample data is input into a pre-constructed classifier initial model, and the classifier initial model is iteratively trained using a stochastic gradient descent algorithm until a preset loss function converges to obtain the classifier model.

[0023] As a further technical solution, the training method further includes:

[0024] Acquire the pre-collected DNA methylation raw sample data, and assign a label to each of the DNA methylation raw sample data; wherein each of the DNA methylation raw sample data includes a sample size, DNA methylation raw data, and DNA unmethylation raw data;

[0025] Preprocessing the DNA methylation raw sample data according to a preset preprocessing rule to obtain the DNA methylation sample data;

[0026] Wherein, if the DNA methylation data is DNA methylation data of a DNA methylation chip, the preprocessing rules at least include: fluorescence correction rules, signal background correction rules, probe filtering rules and signal intensity standardization rules.

[0027] As a further technical solution, the fluorescence correction rule is: respectively calculating the average signal intensity values ​​of the DNA methylation probe set and the DNA unmethylation probe set; respectively calculating the DNA methylation normalization factor and the DNA unmethylation normalization factor based on the average signal intensity values; using the DNA methylation normalization factor and the DNA unmethylation normalization factor to normalize the DNA methylation raw data and the DNA unmethylation raw data in the DNA methylation raw sample data;

[0028] As a further technical solution, the signal background correction rule is: the signal intensity value of the DNA methylation probe and the signal intensity value of the DNA unmethylation probe are respectively subtracted from the DNA methylation and DNA unmethylation signal intensity values ​​of the 5% percentile probes in the negative control probe set arranged from small to large signal intensity, and the minimum signal value is set to 0;

[0029] The probe filtering rules are as follows: filter the probes located on the sex chromosomes; filter the probes with SNP sites within 5 bp of the target CpG site; filter the hybridization non-unique probes; if the DNA methylation data is derived from the GPL13534 chip, the GPL33022 chip and the GPL21145 chip, filter out the probes that are not included in the GPL13534, GPL33022 chip and GPL21145 chip at the same time;

[0030] The signal intensity standardization rule is: make the standard deviation of the signal intensity of all detection probes of each preset type of sample be 1 and the average value be 0.

[0031] As a further technical solution, the training method further includes:

[0032] Acquire a pre-constructed validation set, wherein the validation set includes DNA methylation sample data of each preset type; and the DNA methylation sample data carries a label corresponding to the preset type;

[0033] Inputting the DNA methylation sample data included in the validation set into the trained classifier model to obtain a classification result of the classifier model;

[0034] Counting the classification results, calculating an evaluation index based on the statistical classification results, and evaluating the trained classifier model based on the evaluation index;

[0035] The evaluation index includes at least one of the following indicators:

[0036] Comprehensive accuracy: the ratio of the sample size of the DNA methylation sample data that is correctly classified to the total sample size;

[0037] Recall rate of each preset type: for each preset type, the proportion of the sample volume of the DNA methylation sample data that is actually correctly classified to the required sample volume is counted;

[0038] The accuracy rate of each preset type: for each preset type, the ratio of the sample volume of the DNA methylation sample data that is actually correctly classified to the total volume of all the DNA methylation sample data classified into the preset type is counted;

[0039] AUC value: The DNA methylation sample data that is correctly classified is configured as a positive label, and the DNA methylation sample data that is incorrectly classified is configured as a negative label to calculate the AUC value.

[0040] As a further technical solution, the step of determining the type of DNA methylation data of the sample to be tested based on the classification result includes:

[0041] The preset type corresponding to the maximum probability in the classification results is selected as the type to which the DNA methylation data of the sample to be tested belongs.

[0042] As a further technical solution, the step of inputting the DNA methylation data of the sample to be tested into a pre-trained classifier model comprises:

[0043] The DNA methylation data of the sample to be tested is preprocessed according to the preprocessing rules, and the preprocessed DNA methylation data is input into a pre-trained classifier model.

[0044] In a second aspect, the present invention provides the use of the above-mentioned classification device in the preparation of a product for diagnosing central nervous system tumor types.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] The training samples used in the classification device for central nervous system tumors provided by the present invention are more complete, covering 148 types of CNS tumors, increasing the diversity of tumor methylation, and ensuring the classification accuracy in practical applications; solving the problem that tumor diagnosis results cannot be made at the morphological level, avoiding human subjectivity, and improving diagnostic efficiency; the results of 3-fold cross-validation show that the classifier of the present invention has high accuracy: the accuracy can reach 0.988, and the AUC value is 0.971; it can be used for NGS (second-generation sequencing) data. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 The structure of the classification module in Embodiment 1 of the present invention;

[0049] Figure 2 The proportion of correctly classified samples to all samples;

[0050] Figure 3 The results for correctly classifying the methylation subtypes of central nervous system tumors were similar to those for the first one;

[0051] Figure 4 The results for correctly classifying the methylation subtypes of central nervous system tumors were similar to those for the second subtype;

[0052] Figure 5 The results for correctly classifying the methylation subtypes of central nervous system tumors were similar to those of the third group;

[0053] Figure 6 The results for correctly classifying the CNS tumor methylation subtypes were similar to those for the fourth subtype;

[0054] Figure 7 The recall and precision results for each subtype are as follows;

[0055] Figure 8 The recall and precision results for each subtype are shown in Figure 2;

[0056] Fig. 9 The recall and precision results for each subtype are three;

[0057] Fig.10 The recall and precision results for each subtype are four;

[0058] Fig.11 The receiver operating characteristic curve. DETAILED DESCRIPTION

[0059] The embodiments of the present invention will be described in detail below in conjunction with the embodiments and examples, but it will be appreciated by those skilled in the art that the following embodiments and examples are only used to illustrate the present invention and should not be considered as limiting the scope of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative work are within the scope of protection of the present invention. If specific conditions are not specified, proceed according to normal conditions or conditions recommended by the manufacturer. If the manufacturer is not specified for the reagents or instruments used, they are all conventional products that can be purchased commercially.

[0060] In a first aspect, the present invention provides a classification device for central nervous system tumors, the classification device comprising:

[0061] An acquisition module, used to acquire DNA methylation data of a central nervous system tumor sample to be tested, wherein the DNA methylation data includes a three-dimensional matrix consisting of a sample size, a DNA methylation signal intensity, and a DNA unmethylation signal intensity of the sample to be tested;

[0062] A classification module, used for inputting the DNA methylation data into a pre-trained classifier model, classifying the DNA methylation data through the classifier model, and outputting the classification result of the DNA methylation data; wherein the classifier model is trained based on DNA methylation sample data of multiple preset types, and the classification result includes the probability that the DNA methylation data of the sample to be tested belongs to each preset type;

[0063] A determination module, used to determine the type of DNA methylation data of the sample to be tested based on the classification result;

[0064] The preset types include: "Astroblastoma with MN1 alteration, MN1-BEND2 fusion type", "Angiocentric glioma with MYB / MYBL1 alteration", "Astrocytoma, IDH mutant, high grade", "Astrocytoma, IDH mutant, low grade", "Anaplastic neuroepithelial tumor with condensed nuclei", "Rhabdomyosarcoma, alveolar subtype", "Atypical teratoid / rhabdoid tumor, MYC subtype", "Atypical teratoid / rhabdoid tumor, SHH subtype", "Atypical teratoid / rhabdoid tumor, TYR subtype", "Cauda equina neuroendocrine tumor or paraganglioma, non-CIMP subtype", "Chordoid glioma, PRKCA mutations,” “chordoma,” “central neurocytoma,” “CNS tumors with BCOR / BCORL1 fusions,” “CNS tumors with BCOR internal tandem duplications,” “CNS neuroblastoma, FOXR2 activation,” “CIC rearranged sarcoma,” “primary intracranial sarcoma, DICER1 mutation,” “CNS schwannoma, VGLL fusion,” “choroid plexus carcinoma,” “ameloblastic craniopharyngioma,” “papillary craniopharyngioma,” “choroid plexus papilloma,” “cribriform neuroepithelial tumor, with SMARCB1 alterations,” “diffuse glioneuronal neoplasm with oligodendroglioma-like features and nuclear clusters (DGONC),” “diffuse hemispheric glioma, H3 G34 mutant", "desmoplastic infantile ganglioglioma / desmoplastic infantile astrocytoma", "CNS diffuse large B-cell lymphoma", "diffuse leptomeningeal glioneuronal tumor, subtype 1", "diffuse leptomeningeal glioneuronal tumor, subtype 2", "diffuse midline glioma, H3 K27 alteration, EGFR altered subtype", "diffuse midline glioma, H3 K27 alteration, H3 K27 mutation or EZHIP expression subtypes,” “dysembryoplastic neuroepithelial tumor,” “myxopapillary ependymoma,” “posterior fossa type A ependymoma,” “posterior fossa type B (PFB) ependymoma,” “subependymoma, posterior fossa,” “spinal cord ependymoma,” “spinal cord ependymoma, MYCN amplification,” “subependymoma, spinal cord,” “supratentorial subependymoma,” “supratentorial ependymoma, ZFTA fusion-positive,” “supratentorial ependymoma, ZFTA fusion-positive, ZFTA-RELA fusion subtype, A subtype” ,” “Supratentorial ependymoma, ZFTA fusion-positive, ZFTA-RELA fusion subtype, subtype B,” “Supratentorial ependymoma, YAP1 fusion-positive,” “Rhabdomyosarcoma, embryonal subtype,” “CNS embryonal tumor with BRD4:LEUTX fusion,” “Embryonal tumor with multilayered rosettes, without C19MC alterations,” “Embryonal tumor with multilayered rosettes, C19MC alterations,” “CNS embryonal tumor with PLAG family amplification,” “Extraventricular neuroblastoma,” “Ewing sarcoma,” “Glioblastoma,IDH wild-type, posterior fossa subtype", "glioblastoma, IDH wild-type, mesenchymal subtype, subtype B", "glioblastoma, IDH wild-type, mesenchymal subtype", "glioblastoma, IDH wild-type, with primitive neuronal component", "glioblastoma, IDH wild-type, RTK1 subtype", "glioblastoma, IDH wild-type, RTK2 subtype", "germ cell tumor, KIT wild-type subtype", "germ cell tumor, KIT mutant subtype", "teratoma", "yolk sac tumor", "ganglioglioma", "glioneuronal tumor, subtype A", "with High-grade astrocytoma with lentiform features", "Adult-type diffuse high-grade glioma, IDH wild-type, subtype B", "Adult-type diffuse high-grade glioma, IDH wild-type, subtype E", "Adult-type diffuse high-grade glioma, IDH wild-type, subtype F", "Hemangioblastoma", "Infantile hemispheric glioma", "Inflammatory microenvironment", "Intraocular medulloepithelioma", "Langerhans cell histiocytosis", "Diffuse astrocytoma, MYB or MYBL1 alterations", "Cerebellar liponeurocytoma", "Medulloblastoma, non-WNT / non-SHH, Group 3 Subtype, Subtype I", "Medulloblastoma, non-WNT / non-SHH, Group 3 Subtype, Subtype II", "Medulloblastoma, non-WNT / non-SHH, Group 3 Subtype, Subtype III", "Medulloblastoma, non-WNT / non-SHH, Group 3 Subtype, Subtype IV", "Medulloblastoma, non-WNT / non-SHH, Group 4 Subtype, Subtype V", "Medulloblastoma, non-WNT / non-SHH, Group 4 Subtype, Subtype VI", "Medulloblastoma, non-WNT / non-SHH, Group 4 Subtype, Subtype VII", "Medulloblastoma, non-WNT / non-SHH, Group 4 subtypes, subtype VIII", "Medulloblastoma", "Medulloblastoma, SHH activated, subtype 1", "Medulloblastoma, SHH activated, subtype 2", "Medulloblastoma, SHH activated, subtype 3", "Medulloblastoma, SHH activated, subtype 4", "Medulloblastoma, SHH activated, IDH mutation", "Medulloblastoma, WNT activation", "Melanocytic tumor", "Melanoma metastasis", "Malignant melanotic nerve sheath tumor", "Meningioma, benign", "Meningioma, subtype intermediate", "Meningioma, malignant subtype", "Meningioma, clear cell subtype, SMARCE1 alteration", "Atypical malignant peripheral nerve sheath tumor", "Malignant peripheral nerve sheath tumor", "Myxoid glioneuronal tumor, PDGFRA mutation", "Neuroblastoma, MYCN type", "Neuroblastoma, TMM negative", "Neuroblastoma, ALT / TERT TMM positive", "Neuroepithelial tumor,MN1: CXXC5 fusion", "Neuroepithelial neoplasm with PATZ1 fusion", "Neuroepithelial neoplasm, PLAGL1 fusion", "Plexiform neurofibroma", "Oligodendroglioblastoma, IDH mutation and 1p / 19q co-deletion", "Oligosarcoma, IDH mutant", "Osthoneuroblastoma, IDH wild-type", "Pilocytic astrocytoma, hemispheric", "Pilocytic astrocytoma, infratentorial", "Pilocytic astrocytoma, infratentorial, FGFR1 altered subtype", "Pilocytic astrocytoma, midline", "Pineoblastoma, MYC / FOXR2 activation", "Pineoblastoma tumor, miRNA processing alteration 1 subtype, subtype A", "Pineoblastoma, miRNA processing alteration 1 subtype, subtype B", "Pineoblastoma, miRNA processing alteration 2 subtype", "Diffuse childhood high-grade glioma, H3 and IDH wild-type, subtype A", "Diffuse childhood high-grade glioma, H3 and IDH wild-type, subtype B", "Diffuse childhood high-grade glioma, MYCN subtype", "Diffuse childhood high-grade glioma, RTK subtype", "Papillary glioneuronal tumor, PRKDC fusion", "Pineocytoma", "Pineoblastoma, R B1 change subtype (pineal retinoblastoma)", "Pituitary adenoma, ACTH-secreting subtype", "Pituitary adenoma, gonadotropin-secreting subtype", "Pituitary adenoma, prolactin-secreting subtype", "Pituitary adenoma, STH-secreting subtype, dense granular A subtype", "Pituitary adenoma, STH-secreting subtype, dense granular B subtype", "Pituitary adenoma, STH-secreting subtype, sparse granular subtype", "Pituitary adenoma, TSH-secreting subtype", "Pituitary cell tumor, sellar granular cell tumor, spindle oncocytoma", "CNS plasmacytoma", "Pleomorphic low-grade neuroepithelial tumor of the young", "Moderately differentiated pineal parenchyma Tumor, KBTBD4 alteration, subtype A", "Moderately differentiated pineal parenchymal tumor, KBTBD4 alteration, subtype B", "Pinatal papillary tumor, subtype A", "Pinatal papillary tumor, subtype B", "Pleomorphic xanthoastrocytoma", "Retinoblastoma", "Retinoblastoma, MYCN altered subtype", "Rosette-forming glioneuronal tumor", "Rhabdomyosarcoma, MYOD1 mutant", "Neurilemma", "Subependymal giant cell astrocytoma, TSC1 / TSC2 alteration", "Solitary fibrous tumor", and "Undifferentiated sinonasal carcinoma, IDH2 mutation".

[0065] The training samples used in the classification device for central nervous system tumors provided by the present invention are more complete, covering 148 types of CNS tumors, increasing the diversity of tumor methylation, and ensuring the classification accuracy in practical applications; solving the problem that tumor diagnosis results cannot be made at the morphological level, avoiding human subjectivity, and improving diagnostic efficiency; the results of 3-fold cross-validation show that the classifier of the present invention has high accuracy: the accuracy can reach 0.988, and the AUC value is 0.971; it can be used for NGS (second-generation sequencing) data.

[0066] In some optional embodiments, similar subtypes are merged into thirty-four major categories (preset type categories) based on the similarity of tumor subtypes: "B cell tumors" include "CNS diffuse large B cell lymphoma" and "CNS plasmacytoma"; "CNS germ cell tumors" include "germ cell tumors, KIT wild-type subtype", "germ cell tumors, KIT mutation subtype", "teratomas" and "yolk sac tumors"; "CNS tumors with BCOR mutations" include "CNS tumors with BCOR / BCORL1 fusion" and "CNS tumors with BCOR internal tandem duplications"; "intermediately differentiated pineal parenchymal tumors" include "moderately differentiated pineal parenchymal tumors, KBTBD4 changes, subtype A" and "moderately differentiated pineal parenchymal tumors, KBTBD4 changes, subtype B"; "embryonic tumors with multi-layered rosettes" include "embryonic tumors with multi-layered rosettes, without C19MC changes" and "embryonic tumors with multi-layered rosettes, C19MC changes"; "low-grade glial neuronal tumors" include “Diffuse glioneuronal neoplasm with oligodendroglioma-like features and nuclear clusters (DGONC)”, “dysembryoplastic neuroepithelial tumor”, “extraventricular neurocytoma”, “myxoid glioneuronal neoplasm, PDGFRA gene mutation”, “papillary glioneuronal neoplasm, PRKDC fusion”, and “rosette-forming glioneuronal neoplasm”; “low-grade ganglion cell / neuroepithelial tumor” including “desmoplastic infantile ganglioglioma / desmoplastic infantile astrocytoma”, “ganglioglioma”, “pilocytic astrocytoma”, “ "Pituitary adenoma" includes "pituitary adenoma, ACTH-secreting subtype", "pituitary adenoma, gonadotropin-secreting subtype", "pituitary adenoma, prolactin-secreting subtype", "pituitary adenoma, STH-secreting subtype", "pituitary adenoma, gonadotropin-secreting subtype", "pituitary adenoma, prolactin-secreting subtype", "pituitary adenoma, STH-secreting subtype". "Secreting subtype, dense granular A subtype", "Pituitary adenoma, STH secreting subtype, dense granular B subtype", "Pituitary adenoma, STH secreting subtype, sparse granular subtype" and "Pituitary adenoma, TSH secreting subtype"; "Supratentorial ependymoma, ZFTA fusion positive" includes "Supratentorial ependymoma, ZFTA fusion positive", "Supratentorial ependymoma, ZFTA fusion positive, ZFTA-RELA fusion subtype, subtype A" and "Supratentorial ependymoma, ZFTA fusion positive, ZFTA-RELA fusion subtype, subtype B"; "Diffuse midline glioma, H3 K27 alteration" includes "Diffuse midline glioma, H3 K27 alteration, EGFR alteration subtype" and "Diffuse midline glioma, H3 K27 alteration, H3 K27 mutation or EZHIP expression subtype";“Diffuse childhood high-grade glioma, H3 wild-type and IDH wild-type” includes “Diffuse childhood high-grade glioma, RTK subtype”, “Diffuse hemispheric glioma, H3 G34 mutant”, “Diffuse childhood high-grade glioma, H3 and IDH wild-type, subtype A”, “Diffuse childhood high-grade glioma, H3 and IDH wild-type, subtype B”, and “Diffuse childhood high-grade glioma, MYCN subtype”; “Diffuse glioma, IDH mutant” includes “Astrocytoma, IDH mutant, high grade”, “Astrocytoma, IDH mutant, low grade”, “Oligodendrocyte glioma, IDH mutant and 1p / 19q co-deletion”, and “Oligosarcoma, IDH mutant”; “Diffuse glioma, MYB or MYBL1 mutant” includes “Angiocentric glioma, with MYB / MYBL1 alterations” and “diffuse astrocytoma, with MYB or MYBL1 alterations”; “diffuse glioneuronal tumors” include “diffuse leptomeningeal glioneuronal tumor, subtype 1”, “diffuse leptomeningeal glioneuronal tumor, subtype 2”, and “glioneuronal tumor, subtype A”; “diffuse high-grade neuroepithelial tumor [adult type]” include “anaplastic neuroepithelial tumor with condensed nuclei”, “adult type diffuse high-grade glioma, IDH wild type, subtype B”, “adult type diffuse high-grade glioma, IDH wild type, subtype E”. ” and “adult-type diffuse high-grade glioma, IDH wild-type, subtype F”; “malignant peripheral nerve sheath tumor” includes “atypical malignant peripheral nerve sheath tumor” and “malignant peripheral nerve sheath tumor”; “Schwann cell tumor” includes “CNS nerve sheath tumor, VGLL fusion” and “nerve sheath tumor”; “pineal / retinal RB1-related tumors” include “pinealoblastoma, RB1-altered subtype (pineal retinoblastoma)” and “retinoblastoma”; “pineal papillary tumor” includes “pineal region papillary tumor, subtype A” and “pineal region papillary tumor, subtype B”; “pineal "Pineoblastoma" includes "Pineoblastoma, miRNA processing alteration 1 subtype, subtype A", "Pineoblastoma, miRNA processing alteration 1 subtype, subtype B", and "Pineoblastoma, miRNA processing alteration 2 subtype"; "Pilocytic astrocytoma" includes "Pilocytic astrocytoma, infratentorial", "Pilocytic astrocytoma, infratentorial, FGFR1 altered subtype", and "Pilocytic astrocytoma, midline"; "Neuroblastoma" includes "Neuroblastoma, MYCN type", "Neuroblastoma, TMM negative", and "Neuroblastoma, ALT / TERT TMM positive"; "Embryonal, MYOD1 mutant rhabdomyosarcoma" includes "Rhabdomyosarcoma, embryonal subtype" and "Rhabdomyosarcoma, MYOD1 mutant";“Glioblastoma, IDH wild-type” includes “Glioblastoma, IDH wild-type, posterior fossa subtype”, “Glioblastoma, IDH wild-type, mesenchymal subtype, subtype B”, “Glioblastoma, IDH wild-type, mesenchymal subtype”, “Glioblastoma, IDH wild-type, with primitive neuronal component”, “Glioblastoma, IDH wild-type, RTK1 subtype”, and “Glioblastoma, IDH wild-type, RTK2 subtype”; “Chroid plexus tumor” includes “Chroid plexus carcinoma” and “Chroid plexus papilloma”; “Spinal ependymoma” includes “Spinal ependymoma” and “Spinal ependymoma, MYCN amplification"; "meningioma" includes "meningioma, benign", "meningioma, subtype intermediate", "meningioma, malignant subtype", and "meningioma, clear cell subtype, SMARCE1 alteration"; "atypical teratoid / rhabdoid tumor" includes "atypical teratoid / rhabdoid tumor, MYC subtype", "atypical teratoid / rhabdoid tumor, SHH subtype", and "atypical teratoid / rhabdoid tumor, TYR subtype"; "craniopharyngioma" includes "ameloblastic craniopharyngioma" and "papillary craniopharyngioma"; "cranial nerve and paraspinal nerve tumors" include "malignant melanotic schwannoma" and "plexiform neurofibroma"; “Medulloblastoma, SHH activated” includes “Medulloblastoma, SHH activated, subtype 1”, “Medulloblastoma, SHH activated, subtype 2”, “Medulloblastoma, SHH activated, subtype 3”, “Medulloblastoma, SHH activated, subtype 4” and “Medulloblastoma, SHH activated, IDH mutated”; “Medulloblastoma, non-WNT / non-SHH activated” includes “Medulloblastoma, non-WNT / non-SHH, Group 3 subtype, subtype I”, “Medulloblastoma, non-WNT / non-SHH, Group 3 subtype, subtype II”, “Medulloblastoma, non-WNT / non-SHH, Group 3 subtype, subtype III”, “Medulloblastoma, non-WNT / non-SHH, Group 3 subtype, subtype IV”, “Medulloblastoma, non-WNT / non-SHH, Group 4 subtype, subtype V”, “Medulloblastoma, non-WNT / non-SHH, Group Group 4 subtype, subtype VI", "Medulloblastoma, non-WNT / non-SHH, Group 4 subtype, subtype VII", and "Medulloblastoma, non-WNT / non-SHH, Group 4 subtype, subtype VIII".;

[0067] In some optional embodiments, the DNA methylation data includes DNA methylation data of a DNA methylation chip or second-generation sequencing data;

[0068] The DNA methylation chip includes but is not limited to at least one of the GPL13534, GPL33022 and GPL21145 chips.

[0069] GPL13534, GPL33022 chips and GPL21145 are all CpG site detection technologies in chip sequencing, which can detect a large number of CpG sites in the human genome. For the CpG sites to be detected, they all have type I probes and type II probes. Type I probes have two types of magnetic beads, one for detecting methylated signals and the other for detecting unmethylated signals; while type II probes have only one type of magnetic beads, which can detect methylated signals and unmethylated signals. Methylated signals and unmethylated signals are reflected in green and red in color, respectively. When the hybridization reaction between the probe and the sample in the chip is completed, the color (fluorescence) signal intensity can be read by the iScan and HiScan®SQ scanning systems.

[0070] In some optional embodiments, the classifier model includes: an input layer, an intermediate layer, and an output layer;

[0071] The intermediate layer includes a preset number of convolutional layers, a pooling layer between two adjacent convolutional layers, a flattening layer connected to the last convolutional layer, and a preset number of fully connected layers;

[0072] The number of neurons configured in the output layer is consistent with the number of the preset types.

[0073] In some optional embodiments, the classifier includes 1 input layer, 3 convolutional layers, 2 pooling layers, 1 flattening layer, 5 fully connected layers and 1 output layer.

[0074] In some optional embodiments, the training method of the classifier model includes:

[0075] Obtaining a pre-constructed training set, wherein the training set includes DNA methylation sample data of each preset type; and the DNA methylation sample data carries a label corresponding to the preset type;

[0076] The preset type of DNA methylation sample data is input into a pre-constructed classifier initial model, and the classifier initial model is iteratively trained using a stochastic gradient descent algorithm until a preset loss function converges to obtain the classifier model.

[0077] In some optional embodiments, the label conversion may be, for example: randomly converting the above 148 tumor types into numbers between 0 and 147, with each tumor corresponding to a number, to achieve digital processing of tumor types.

[0078] In some optional implementations, the training method further includes:

[0079] Acquire the pre-collected DNA methylation raw sample data, and assign a label to each of the DNA methylation raw sample data; wherein each of the DNA methylation raw sample data includes a sample size, DNA methylation raw data, and DNA unmethylation raw data;

[0080] Preprocessing the DNA methylation raw sample data according to a preset preprocessing rule to obtain the DNA methylation sample data;

[0081] Wherein, if the DNA methylation data is DNA methylation data of a DNA methylation chip, the preprocessing rules at least include: fluorescence correction rules, signal background correction rules, probe filtering rules and signal intensity standardization rules.

[0082] In some optional embodiments, the fluorescence correction rule is: respectively calculate the average signal intensity values ​​of the DNA methylation probe set and the DNA unmethylation probe set; respectively calculate the DNA methylation normalization factor and the DNA unmethylation normalization factor based on the average signal intensity values; and use the DNA methylation normalization factor and the DNA unmethylation normalization factor to standardize the DNA methylation raw data and the DNA unmethylation raw data in the DNA methylation raw sample data.

[0083] It should be noted that the average signal intensity value of the DNA methylation probe set refers to the sum of the green fluorescence signal intensities divided by the number of corrected green fluorescence probes; the average signal intensity value of the DNA unmethylation probe set refers to the sum of the red fluorescence signal intensities divided by the number of corrected red fluorescence probes.

[0084] Green fluorescence correction factor: F 绿 =(§ 绿 +§ 红 )×1 / 2×1 / (§ 绿 );

[0085] Red fluorescence correction factor: F 红 =(§ 绿 +§ 红 )×1 / 2×1 / (§ 红 ).

[0086] Among them, F 绿 is the green fluorescence correction factor, F 红 is the red fluorescence correction factor, § 绿 is the average signal intensity value of the DNA methylation probe set, 红 is the average signal intensity value of the DNA unmethylated probe set.

[0087] The fluorescence correction method is: multiply the green fluorescence signal intensity by F 绿; Red fluorescence signal intensity multiplied by F 红 .

[0088] In some optional embodiments, the signal background correction rule is: the signal intensity value of the DNA methylation probe and the signal intensity value of the DNA unmethylation probe are respectively subtracted from the DNA methylation and DNA unmethylation signal intensity values ​​of the 5% percentile probes in the negative control probe set arranged from small to large signal intensity, and the minimum signal value is set to 0;

[0089] In some optional embodiments, the probe filtering rules are: filter probes located on sex chromosomes; filter probes with SNP sites within 5 bp of the target CpG site; filter hybridization non-unique probes; if the DNA methylation data is derived from the GPL13534 chip, the GPL33022 chip and the GPL21145 chip, filter out probes that are not simultaneously included in the GPL13534, GPL33022 chip and the GPL21145 chip. The filtering rules are used to improve the accuracy of the classification results.

[0090] In some optional embodiments, the signal intensity normalization rule is: the standard deviation of the signal intensity of all detection probes of each preset type of sample is 1, and the average value is 0.

[0091] The general process is, for example, first calculating the standard deviation and average value of the signal intensities of all probes of each sample, and then subtracting the average value from the signal intensity of each probe and dividing the result by the standard deviation.

[0092] In some optional implementations, the training method further includes:

[0093] Acquire a pre-constructed validation set, wherein the validation set includes DNA methylation sample data of each preset type; and the DNA methylation sample data carries a label corresponding to the preset type;

[0094] Inputting the DNA methylation sample data included in the validation set into the trained classifier model to obtain a classification result of the classifier model;

[0095] Counting the classification results, calculating an evaluation index based on the statistical classification results, and evaluating the trained classifier model based on the evaluation index;

[0096] The evaluation index includes at least one of the following indicators:

[0097] Comprehensive accuracy: the ratio of the sample size of the DNA methylation sample data that is correctly classified to the total sample size;

[0098] Recall rate of each preset type: for each preset type, the proportion of the sample volume of the DNA methylation sample data that is actually correctly classified to the required sample volume is counted;

[0099] The accuracy rate of each preset type: for each preset type, the ratio of the sample volume of the DNA methylation sample data that is actually correctly classified to the total volume of all the DNA methylation sample data classified into the preset type is counted;

[0100] AUC value: The DNA methylation sample data that is correctly classified is configured as a positive label, and the DNA methylation sample data that is incorrectly classified is configured as a negative label to calculate the AUC value.

[0101] In some optional embodiments, the step of determining the type to which the DNA methylation data of the sample to be tested belongs based on the classification result includes:

[0102] The preset type corresponding to the maximum probability in the classification results is selected as the type to which the DNA methylation data of the sample to be tested belongs.

[0103] In some optional embodiments, the step of determining the type of the DNA methylation data of the sample to be tested based on the classification result further includes:

[0104] The preset category with the highest probability in the classification results is selected as the type to which the DNA methylation data of the sample to be tested belongs.

[0105] In some optional embodiments, the step of inputting the DNA methylation data of the sample to be tested into a pre-trained classifier model comprises:

[0106] The DNA methylation data of the sample to be tested is preprocessed according to the preprocessing rules, and the preprocessed DNA methylation data is input into a pre-trained classifier model.

[0107] Preprocessing is performed to improve the accuracy of classification results.

[0108] In a second aspect, the present invention provides the use of the above-mentioned classification device in the preparation of a product for diagnosing central nervous system tumor types.

[0109] The classification device provided by the present invention can realize accurate classification of 148 types of central nervous system tumors and can be used to prepare products for diagnosing central nervous system tumor types.

[0110] The present invention is further described below by means of specific examples and comparative examples. However, it should be understood that these examples are only used for more detailed description and should not be construed as limiting the present invention in any form.

[0111] Example 1

[0112] A classification device for central nervous system tumors, comprising:

[0113] An acquisition module, used to acquire DNA methylation data of a central nervous system tumor sample to be tested, wherein the DNA methylation data includes a three-dimensional matrix consisting of a sample size, a DNA methylation signal intensity, and a DNA unmethylation signal intensity of the sample to be tested;

[0114] A classification module, used for inputting the DNA methylation data into a pre-trained classifier model, classifying the DNA methylation data through the classifier model, and outputting the classification result of the DNA methylation data; wherein the classifier model is trained based on DNA methylation sample data of multiple preset types, and the classification result includes the probability that the DNA methylation data of the sample to be tested belongs to each preset type;

[0115] A determination module, used to determine the type of DNA methylation data of the sample to be tested based on the classification result;

[0116] Among them, the preset types include: "Astroblastoma with MN1 alteration, MN1-BEND2 fusion type", "Angiocentric glioma with MYB / MYBL1 alteration", "Astrocytoma, IDH mutant, high grade", "Astrocytoma, IDH mutant, low grade", "Anaplastic neuroepithelial tumor with condensed nuclei", "Rhabdomyosarcoma, alveolar subtype", "Atypical teratoid / rhabdoid tumor, MYC subtype", "Atypical teratoid / rhabdoid tumor, SHH subtype", "Atypical teratoid / rhabdoid tumor, TYR subtype", "Cauda equina neuroendocrine tumor or paraganglioma, non-CIMP subtype", "Chordoid glioma, PRKCA mutant", "Chordoma", "Central neurocytoma", "CNS tumor with BCOR / BCORL1 fusion", "CNS tumor with BCOR internal tandem duplication" ,” “CNS neuroblastoma, FOXR2 activated,” “CIC rearranged sarcoma,” “Primary intracranial sarcoma, DICER1 mutation,” “CNS schwannoma, VGLL fusion,” “Chroid plexus carcinoma,” “Ameloblastic craniopharyngioma,” “Papillary craniopharyngioma,” “Choroid plexus papilloma,” “Cribriform neuroepithelial tumor, with SMARCB1 alterations,” “Diffuse glioneuronal neoplasm with oligodendroglioma-like features and nuclear clusters (DGONC),” “Diffuse hemispheric glioma, H3G34 mutant,” “Desmoplastic infantile ganglioglioma / desmoplastic infantile astrocytoma,” “CNS diffuse large B-cell lymphoma,” “Diffuse leptomeningeal glioneuronal tumor, subtype 1,” “Diffuse leptomeningeal glioneuronal tumor, subtype 2,” “Diffuse midline glioma, H3 K27 alterations, EGFR altered subtype,” “Diffuse midline glioma, H3 K27 alterations, H3 K27 mutation or EZHIP expression subtypes,” “dysembryoplastic neuroepithelial tumor,” “myxopapillary ependymoma,” “posterior fossa type A ependymoma,” “posterior fossa type B (PFB) ependymoma,” “subependymoma, posterior fossa,” “spinal cord ependymoma,” “spinal cord ependymoma, MYCN amplification,” “subependymoma, spinal cord,” “supratentorial subependymoma,” “supratentorial ependymoma, ZFTA fusion-positive,” “supratentorial ependymoma, ZFTA fusion-positive, ZFTA-RELA fusion subtype, A subtype” ,” “Supratentorial ependymoma, ZFTA fusion-positive, ZFTA-RELA fusion subtype, subtype B,” “Supratentorial ependymoma, YAP1 fusion-positive,” “Rhabdomyosarcoma, embryonal subtype,” “CNS embryonal tumor with BRD4:LEUTX fusion,” “Embryonal tumor with multilayered rosettes, without C19MC alterations,” “Embryonal tumor with multilayered rosettes, C19MC alterations,” “CNS embryonal tumor with PLAG family amplification,” “Extraventricular neuroblastoma,” “Ewing sarcoma,” “Glioblastoma,IDH wild-type, posterior fossa subtype", "glioblastoma, IDH wild-type, mesenchymal subtype, subtype B", "glioblastoma, IDH wild-type, mesenchymal subtype", "glioblastoma, IDH wild-type, with primitive neuronal component", "glioblastoma, IDH wild-type, RTK1 subtype", "glioblastoma, IDH wild-type, RTK2 subtype", "germ cell tumor, KIT wild-type subtype", "germ cell tumor, KIT mutant subtype", "teratoma", "yolk sac tumor", "ganglioglioma", "glioneuronal tumor, subtype A", "with High-grade astrocytoma with lentiform features", "Adult-type diffuse high-grade glioma, IDH wild-type, subtype B", "Adult-type diffuse high-grade glioma, IDH wild-type, subtype E", "Adult-type diffuse high-grade glioma, IDH wild-type, subtype F", "Hemangioblastoma", "Infantile hemispheric glioma", "Inflammatory microenvironment", "Intraocular medulloepithelioma", "Langerhans cell histiocytosis", "Diffuse astrocytoma, MYB or MYBL1 alterations", "Cerebellar liponeurocytoma", "Medulloblastoma, non-WNT / non-SHH, Group 3 Subtype, Subtype I", "Medulloblastoma, non-WNT / non-SHH, Group 3 Subtype, Subtype II", "Medulloblastoma, non-WNT / non-SHH, Group 3 Subtype, Subtype III", "Medulloblastoma, non-WNT / non-SHH, Group 3 Subtype, Subtype IV", "Medulloblastoma, non-WNT / non-SHH, Group 4 Subtype, Subtype V", "Medulloblastoma, non-WNT / non-SHH, Group 4 Subtype, Subtype VI", "Medulloblastoma, non-WNT / non-SHH, Group 4 Subtype, Subtype VII", "Medulloblastoma, non-WNT / non-SHH, Group 4 subtypes, subtype VIII", "Medulloblastoma", "Medulloblastoma, SHH activated, subtype 1", "Medulloblastoma, SHH activated, subtype 2", "Medulloblastoma, SHH activated, subtype 3", "Medulloblastoma, SHH activated, subtype 4", "Medulloblastoma, SHH activated, IDH mutation", "Medulloblastoma, WNT activation", "Melanocytic tumor", "Melanoma metastasis", "Malignant melanotic nerve sheath tumor", "Meningioma, benign", "Meningioma, subtype intermediate", "Meningioma, malignant subtype", "Meningioma, clear cell subtype, SMARCE1 alteration", "Atypical malignant peripheral nerve sheath tumor", "Malignant peripheral nerve sheath tumor", "Myxoid glioneuronal tumor, PDGFRA mutation", "Neuroblastoma, MYCN type", "Neuroblastoma, TMM negative", "Neuroblastoma, ALT / TERT TMM positive", "Neuroepithelial tumor,MN1: CXXC5 fusion", "Neuroepithelial neoplasm with PATZ1 fusion", "Neuroepithelial neoplasm, PLAGL1 fusion", "Plexiform neurofibroma", "Oligodendroglioblastoma, IDH mutation and 1p / 19q co-deletion", "Oligosarcoma, IDH mutant", "Osthoneuroblastoma, IDH wild-type", "Pilocytic astrocytoma, hemispheric", "Pilocytic astrocytoma, infratentorial", "Pilocytic astrocytoma, infratentorial, FGFR1 altered subtype", "Pilocytic astrocytoma, midline", "Pineoblastoma, MYC / FOXR2 activation", "Pineoblastoma tumor, miRNA processing alteration 1 subtype, subtype A", "Pineoblastoma, miRNA processing alteration 1 subtype, subtype B", "Pineoblastoma, miRNA processing alteration 2 subtype", "Diffuse childhood high-grade glioma, H3 and IDH wild-type, subtype A", "Diffuse childhood high-grade glioma, H3 and IDH wild-type, subtype B", "Diffuse childhood high-grade glioma, MYCN subtype", "Diffuse childhood high-grade glioma, RTK subtype", "Papillary glioneuronal tumor, PRKDC fusion", "Pineocytoma", "Pineoblastoma, R B1 change subtype (pineal retinoblastoma)", "Pituitary adenoma, ACTH-secreting subtype", "Pituitary adenoma, gonadotropin-secreting subtype", "Pituitary adenoma, prolactin-secreting subtype", "Pituitary adenoma, STH-secreting subtype, dense granular A subtype", "Pituitary adenoma, STH-secreting subtype, dense granular B subtype", "Pituitary adenoma, STH-secreting subtype, sparse granular subtype", "Pituitary adenoma, TSH-secreting subtype", "Pituitary cell tumor, sellar granular cell tumor, spindle oncocytoma", "CNS plasmacytoma", "Pleomorphic low-grade neuroepithelial tumor of the young", "Moderately differentiated pineal parenchyma Tumor, KBTBD4 alteration, subtype A", "Moderately differentiated pineal parenchymal tumor, KBTBD4 alteration, subtype B", "Pinatal papillary tumor, subtype A", "Pinatal papillary tumor, subtype B", "Pleomorphic xanthoastrocytoma", "Retinoblastoma", "Retinoblastoma, MYCN altered subtype", "Rosette-forming glioneuronal tumor", "Rhabdomyosarcoma, MYOD1 mutant", "Neurilemma", "Subependymal giant cell astrocytoma, TSC1 / TSC2 alteration", "Solitary fibrous tumor", and "Undifferentiated sinonasal carcinoma, IDH2 mutation".

[0117] According to the similarity of tumor subtypes, similar subtypes are merged into thirty-four major categories (pre-set type categories): "B cell tumors" include "CNS diffuse large B cell lymphoma" and "CNS plasmacytoma"; "CNS germ cell tumors" include "germ cell tumor, KIT wild-type subtype", "germ cell tumor, KIT mutation subtype", "teratoma" and "yolk sac tumor"; "CNS tumor with BCOR mutation" includes "CNS tumor with BCOR / BCORL1 fusion" and "CNS tumor with BCOR internal tandem duplication"; "intermediately differentiated pineal parenchymal tumor" includes "moderately differentiated pineal parenchymal tumor, KBTBD4 change, subtype A" and "moderately differentiated pineal parenchymal tumor, KBTBD4 change, subtype B"; "embryonic tumor with multilayer rosettes" includes "embryonic tumor with multilayer rosettes, no C19MC change" and "embryonic tumor with multilayer rosettes, C19MC change"; "low-grade glioneuronal tumor" includes “Diffuse glioneuronal neoplasm with oligodendroglioma-like features and nuclear clusters (DGONC)”, “dysembryoplastic neuroepithelial tumor”, “extraventricular neurocytoma”, “myxoid glioneuronal neoplasm, PDGFRA gene mutation”, “papillary glioneuronal neoplasm, PRKDC fusion”, and “rosette-forming glioneuronal neoplasm”; “low-grade ganglion cell / neuroepithelial tumor” including “desmoplastic infantile ganglioglioma / desmoplastic infantile astrocytoma”, “ganglioglioma”, “pilocytic astrocytoma”, “ "Pituitary adenoma" includes "pituitary adenoma, ACTH-secreting subtype", "pituitary adenoma, gonadotropin-secreting subtype", "pituitary adenoma, prolactin-secreting subtype", "pituitary adenoma, STH-secreting subtype", "pituitary adenoma, gonadotropin-secreting subtype", "pituitary adenoma, prolactin-secreting subtype", "pituitary adenoma, STH-secreting subtype". "Secreting subtype, dense granular A subtype", "Pituitary adenoma, STH secreting subtype, dense granular B subtype", "Pituitary adenoma, STH secreting subtype, sparse granular subtype" and "Pituitary adenoma, TSH secreting subtype"; "Supratentorial ependymoma, ZFTA fusion positive" includes "Supratentorial ependymoma, ZFTA fusion positive", "Supratentorial ependymoma, ZFTA fusion positive, ZFTA-RELA fusion subtype, subtype A" and "Supratentorial ependymoma, ZFTA fusion positive, ZFTA-RELA fusion subtype, subtype B"; "Diffuse midline glioma, H3 K27 alteration" includes "Diffuse midline glioma, H3 K27 alteration, EGFR alteration subtype" and "Diffuse midline glioma, H3 K27 alteration, H3 K27 mutation or EZHIP expression subtype";“Diffuse childhood high-grade glioma, H3 wild-type and IDH wild-type” includes “Diffuse childhood high-grade glioma, RTK subtype”, “Diffuse hemispheric glioma, H3 G34 mutant”, “Diffuse childhood high-grade glioma, H3 and IDH wild-type, subtype A”, “Diffuse childhood high-grade glioma, H3 and IDH wild-type, subtype B”, and “Diffuse childhood high-grade glioma, MYCN subtype”; “Diffuse glioma, IDH mutant” includes “Astrocytoma, IDH mutant, high grade”, “Astrocytoma, IDH mutant, low grade”, “Oligodendrocyte glioma, IDH mutant and 1p / 19q co-deletion”, and “Oligosarcoma, IDH mutant”; “Diffuse glioma, MYB or MYBL1 mutant” includes “Angiocentric glioma, with MYB / MYBL1 alterations” and “diffuse astrocytoma, with MYB or MYBL1 alterations”; “diffuse glioneuronal tumors” include “diffuse leptomeningeal glioneuronal tumor, subtype 1”, “diffuse leptomeningeal glioneuronal tumor, subtype 2”, and “glioneuronal tumor, subtype A”; “diffuse high-grade neuroepithelial tumor [adult type]” include “anaplastic neuroepithelial tumor with condensed nuclei”, “adult type diffuse high-grade glioma, IDH wild type, subtype B”, “adult type diffuse high-grade glioma, IDH wild type, subtype E”. ” and “adult-type diffuse high-grade glioma, IDH wild-type, subtype F”; “malignant peripheral nerve sheath tumor” includes “atypical malignant peripheral nerve sheath tumor” and “malignant peripheral nerve sheath tumor”; “Schwann cell tumor” includes “CNS nerve sheath tumor, VGLL fusion” and “nerve sheath tumor”; “pineal / retinal RB1-related tumors” include “pinealoblastoma, RB1-altered subtype (pineal retinoblastoma)” and “retinoblastoma”; “pineal papillary tumor” includes “pineal region papillary tumor, subtype A” and “pineal region papillary tumor, subtype B”; “pineal "Pineoblastoma" includes "Pineoblastoma, miRNA processing alteration 1 subtype, subtype A", "Pineoblastoma, miRNA processing alteration 1 subtype, subtype B", and "Pineoblastoma, miRNA processing alteration 2 subtype"; "Pilocytic astrocytoma" includes "Pilocytic astrocytoma, infratentorial", "Pilocytic astrocytoma, infratentorial, FGFR1 altered subtype", and "Pilocytic astrocytoma, midline"; "Neuroblastoma" includes "Neuroblastoma, MYCN type", "Neuroblastoma, TMM negative", and "Neuroblastoma, ALT / TERT TMM positive"; "Embryonal, MYOD1 mutant rhabdomyosarcoma" includes "Rhabdomyosarcoma, embryonal subtype" and "Rhabdomyosarcoma, MYOD1 mutant";“Glioblastoma, IDH wild-type” includes “Glioblastoma, IDH wild-type, posterior fossa subtype”, “Glioblastoma, IDH wild-type, mesenchymal subtype, subtype B”, “Glioblastoma, IDH wild-type, mesenchymal subtype”, “Glioblastoma, IDH wild-type, with primitive neuronal component”, “Glioblastoma, IDH wild-type, RTK1 subtype”, and “Glioblastoma, IDH wild-type, RTK2 subtype”; “Chroid plexus tumor” includes “Chroid plexus carcinoma” and “Chroid plexus papilloma”; “Spinal ependymoma” includes “Spinal ependymoma” and “Spinal ependymoma, MYCN amplification"; "meningioma" includes "meningioma, benign", "meningioma, subtype intermediate", "meningioma, malignant subtype", and "meningioma, clear cell subtype, SMARCE1 alteration"; "atypical teratoid / rhabdoid tumor" includes "atypical teratoid / rhabdoid tumor, MYC subtype", "atypical teratoid / rhabdoid tumor, SHH subtype", and "atypical teratoid / rhabdoid tumor, TYR subtype"; "craniopharyngioma" includes "ameloblastic craniopharyngioma" and "papillary craniopharyngioma"; "cranial nerve and paraspinal nerve tumors" include "malignant melanotic schwannoma" and "plexiform neurofibroma"; “Medulloblastoma, SHH activated” includes “Medulloblastoma, SHH activated, subtype 1”, “Medulloblastoma, SHH activated, subtype 2”, “Medulloblastoma, SHH activated, subtype 3”, “Medulloblastoma, SHH activated, subtype 4” and “Medulloblastoma, SHH activated, IDH mutated”; “Medulloblastoma, non-WNT / non-SHH activated” includes “Medulloblastoma, non-WNT / non-SHH, Group 3 subtype, subtype I”, “Medulloblastoma, non-WNT / non-SHH, Group 3 subtype, subtype II”, “Medulloblastoma, non-WNT / non-SHH, Group3 subtype, subtype III”, “Medulloblastoma, non-WNT / non-SHH, Group 3 subtype, subtype IV”, “Medulloblastoma, non-WNT / non-SHH, Group 4 subtype, subtype V”, “Medulloblastoma, non-WNT / non-SHH, Group Group 4 subtype, subtype VI", "Medulloblastoma, non-WNT / non-SHH, Group 4 subtype, subtype VII", and "Medulloblastoma, non-WNT / non-SHH, Group 4 subtype, subtype VIII".;

[0118] Among them, DNA methylation data needs to be preprocessed according to the following rules:

[0119] 1. Label conversion: The above 148 tumor types are randomly converted into numbers between 0 and 147, with each tumor corresponding to a number.

[0120] 2. Preliminary processing of chip probes:

[0121] 1) Standardize the fluorescence signal intensity between different chips:

[0122] Green and red reference signal intensities: Calculate the average values ​​of the green fluorescence calibration probe set (e.g., consisting of NORM_C and NORM_G probes) and the red fluorescence calibration probe set (e.g., consisting of NORM_A and NORM_T), respectively. §Green (sum of green fluorescence signal intensities divided by the number of calibration green fluorescence probes) and §Red (sum of red fluorescence signal intensities divided by the number of calibration red fluorescence probes);

[0123] Green fluorescence correction factor: F 绿 =(§ 绿 +§ 红 )×1 / 2×1 / (§ 绿 );

[0124] Red fluorescence correction factor: F 红 =(§ 绿 +§ 红 )×1 / 2×1 / (§ 红 ).

[0125] Among them, F 绿 is the green fluorescence correction factor, F 红 is the red fluorescence correction factor, § 绿 is the average signal intensity value of the DNA methylation probe set, 红 is the average signal intensity value of the DNA unmethylated probe set.

[0126] The fluorescence correction method is: multiply the green fluorescence signal intensity by F 绿 ; Red fluorescence signal intensity multiplied by F 红 .

[0127] 2) Red and green fluorescence signal background correction: All green and red fluorescence signal values ​​are subtracted from the DNA methylation and DNA unmethylation signal intensity values ​​of the 5% percentile probes in the negative control probe set (e.g., composed of NEGATIVE probes) whose signal intensity is arranged from small to large, and the minimum signal value is set to 0.

[0128] 3) Probe filtering: filter probes located on sex chromosomes; filter probes with SNP sites within 5bp of the target CpG site; filter hybridization non-unique probes; filter out probes that are not included in both the GPL13534 and GPL21145 chips.

[0129] 4) The signal intensity normalization rule is: the standard deviation of the signal intensity of all detection probes of each preset type of sample is 1 and the average value is 0.

[0130] like Figure 1 As shown in the figure, the classifier model includes 1 input layer, 3 convolutional layers, 2 pooling layers, 1 flattening layer, 5 fully connected layers and 1 output layer. The construction method is as follows:

[0131] 1. Use the preprocessed three-dimensional data as the input layer (sample size, methylation signal intensity, unmethylation signal intensity);

[0132] 2. Layers 2 to 13 are data training layers:

[0133] A) Convolutional layer parameter settings: 32 randomly initialized convolution kernels, convolution kernel length is 3, window sliding length is 1, activation function is ReLu: ;

[0134] B) Pooling layer parameter settings: pooling window length is 3, window sliding length is 1;

[0135] C) Fully connected layer parameter settings: 256 neurons, activation function is ReLu: , an L2 regularization penalty is applied to the weights and the output values ​​of this layer.

[0136] 3. Output layer parameter setting: 148 neurons, using softmax as the activation function of this layer.

[0137] 4. During the training process, the Adam stochastic gradient descent algorithm and the cross entropy loss indicator are used to optimize the weight value.

[0138] The DNA methylation data used in the training process comes from the GEO database (Gene Expression Omnibus), and the sequencing platforms include GPL13534 and GPL21145.

[0139] Classifier performance evaluation:

[0140] 1. Determination of classification threshold: After a sample is sequenced by methylation chip, a three-dimensional matrix of methylation signal intensity and unmethylation signal intensity will be obtained. After the three-dimensional matrix is ​​input into the above-mentioned trained classifier, a tumor classification score (MC score) and a tumor major classification value (MCF score) will be obtained. The score is between [0,1], indicating the probability of the sample being classified into the corresponding tumor. After verification by a large number of samples, the conditions for a sample to be correctly classified are: MC score is not less than 0.9, that is, the result is considered credible; MCF score (sum of multiple MC scores) is not less than 0.9 and the subordinate MC score is not less than 0.5, that is, the result is considered credible (acceptable conclusion using three-fold cross-validation and independent verification).

[0141] 2. Three-fold cross validation and independent validation methods were used to evaluate the classifier performance.

[0142] A) Three-fold cross-validation: The 3344 training samples were divided into three data sets in equal proportions, two of which were used as training data sets and the other as validation data set (i.e., the ratio of training set and validation set samples was 2:1. Each data set included all central nervous system tumor subtypes). This classifier performance evaluation method was iterated three times.

[0143] B) Independent validation: The independent validation dataset is independent of the training dataset used to develop this classifier and does not participate in the training of the classifier. A total of 833 samples (covering some central nervous system tumor subtypes) were selected, of which 817 samples were correctly classified.

[0144] 3. Statistical performance evaluation indicators (results using three-fold cross validation method):

[0145] A) Overall accuracy: the proportion of correctly classified samples to all samples; see Figure 2 , Figure 3 , Figure 4 , Figure 5 and Figure 6 ( Figure 3-Figure 6 The horizontal axis is the maximum similarity score).

[0146] B) Recall rate of each subtype: for each subtype, the proportion of samples that are actually correctly classified to the expected number of samples is counted; see Figure 7 , Figure 8 , Fig. 9 and Fig.10 .

[0147] C) Precision of each subtype: For each subtype, the proportion of samples that are actually correctly classified as the corresponding subtype to the total number of samples classified as the subtype is counted; see Figure 7 , Figure 8 , Fig. 9 and Fig.10 .

[0148] D) Receiver operating characteristic curve (AUC): label all correctly classified samples as positive, label all incorrectly classified samples as negative, and calculate AUC. Fig.11 .

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A classification device for central nervous system tumors, characterized in that: The classification device comprises: An acquisition module, used to acquire DNA methylation data of a central nervous system tumor sample to be tested, wherein the DNA methylation data includes a three-dimensional matrix consisting of a sample size, a DNA methylation signal intensity, and a DNA unmethylation signal intensity of the sample to be tested; A classification module, used for inputting the DNA methylation data into a pre-trained classifier model, classifying the DNA methylation data through the classifier model, and outputting the classification result of the DNA methylation data; wherein the classifier model is trained based on DNA methylation sample data of multiple preset types, and the classification result includes the probability that the DNA methylation data of the sample to be tested belongs to each preset type; The classifier model includes: an input layer, an intermediate layer and an output layer; The intermediate layer includes a preset number of convolutional layers, a pooling layer between two adjacent convolutional layers, a flattening layer connected to the last convolutional layer, and a preset number of fully connected layers; Wherein, the number of neurons configured in the output layer is consistent with the number of the preset types; A determination module, used to determine the type of DNA methylation data of the sample to be tested based on the classification result; The preset types include: "Astroblastoma with MN1 alteration, MN1-BEND2 fusion type", "Angiocentric glioma with MYB / MYBL1 alteration", "Astrocytoma, IDH mutant, high grade", "Astrocytoma, IDH mutant, low grade", "Anaplastic neuroepithelial tumor with condensed nuclei", "Rhabdomyosarcoma, alveolar subtype", "Atypical teratoid / rhabdoid tumor, MYC subtype", "Atypical teratoid / rhabdoid tumor, SHH subtype", "Atypical teratoid / rhabdoid tumor, TYR subtype", "Cauda equina neuroendocrine tumor or paraganglioma, non-CIMP subtype", "Chordoid glioma, P RKCA mutant", "Chordoma", "CNS neurocytoma", "CNS neoplasms with BCOR / BCORL1 fusion", "CNS neoplasms with BCOR internal tandem duplication", "CNS neuroblastoma, FOXR2 activated", "CIC rearranged sarcoma", "primary intracranial sarcoma, DICER1 mutation", "CNS schwannoma, VGLL fusion", "choroid plexus carcinoma", "ameloblastic craniopharyngioma", "papillary craniopharyngioma", "choroid plexus papilloma", "cribriform neuroepithelial tumor with SMARCB1 alterations", "diffuse glioneuronal tumor with oligodendroglioma-like features and nuclear clusters", "diffuse hemispheric glioma, H3 G34 mutant", "desmoplastic infantile ganglioglioma / desmoplastic infantile astrocytoma", "CNS diffuse large B-cell lymphoma", "diffuse leptomeningeal glioneuronal tumor, subtype 1", "diffuse leptomeningeal glioneuronal tumor, subtype 2", "diffuse midline glioma, H3 K27 altered, EGFR altered subtype", "diffuse midline glioma, H3 K27 alterations, H3K27 mutations or EZHIP expression subtypes", "dysembryoplastic neuroepithelial tumors", "myxopapillary ependymoma", "posterior fossa type A ependymoma", "posterior fossa type B ependymoma", "subependymoma, posterior fossa", "spinal ependymoma", "spinal ependymoma, MYCN amplification", "subependymoma, spinal cord", "supratentorial subependymoma", "supratentorial ependymoma, ZFTA fusion-positive", "supratentorial ependymoma, ZFTA fusion-positive, ZFTA-RELA fusion subtype, subtype A", " Supratentorial ependymoma, ZFTA fusion-positive, ZFTA-RELA fusion subtype, subtype B", "Supratentorial ependymoma, YAP1 fusion-positive", "Rhabdomyosarcoma, embryonal subtype", "CNS embryonal tumor with BRD4:LEUTX fusion", "Embryonal tumor with multilayered rosettes, without C19MC alterations", "Embryonal tumor with multilayered rosettes, C19MC alterations", "CNS embryonal tumor with PLAG family amplification", "Extraventricular neuroblastoma", "Ewing sarcoma", "Glioblastoma, IDH wild-type,Posterior fossa subtype", "Glioblastoma, IDH wild-type, mesenchymal subtype, subtype B", "Glioblastoma, IDH wild-type, mesenchymal subtype", "Glioblastoma, IDH wild-type, with primitive neuronal component", "Glioblastoma, IDH wild-type, RTK1 subtype", "Glioblastoma, IDH wild-type, RTK2 subtype", "Germoma, KIT wild-type subtype", "Germoma, KIT mutant subtype", "Teratoma", "Yolk sac tumor", "Ganglioblastoma", "Glioneuronal tumor, subtype A", "With lentiform component", "High-grade astrocytoma with nuclear features", "Adult diffuse high-grade glioma, IDH wild-type, subtype B", "Adult diffuse high-grade glioma, IDH wild-type, subtype E", "Adult diffuse high-grade glioma, IDH wild-type, subtype F", "Hemangioblastoma", "Infantile hemispheric glioma", "Inflammatory microenvironment", "Intraocular medulloepithelioma", "Langerhans cell histiocytosis", "Diffuse astrocytoma, MYB or MYBL1 alterations", "Cerebellar liponeurocytoma", "Medulloblastoma, non-WNT / non-SHH, Group "Medulloblastoma, non-WNT / non-SHH, Group 3 subtype, subtype I", "Medulloblastoma, non-WNT / non-SHH, Group 3 subtype, subtype II", "Medulloblastoma, non-WNT / non-SHH, Group 3 subtype, subtype III", "Medulloblastoma, non-WNT / non-SHH, Group 3 subtype, subtype IV", "Medulloblastoma, non-WNT / non-SHH, Group 4 subtype, subtype V", "Medulloblastoma, non-WNT / non-SHH, Group 4 subtype, subtype VI", "Medulloblastoma, non-WNT / non-SHH, Group 4 subtype, subtype VII", "Medulloblastoma, non-WNT / non-SHH, Group 4, subtype VIII", "Medulloblastoma", "Medulloblastoma, SHH activated, subtype 1", "Medulloblastoma, SHH activated, subtype 2", "Medulloblastoma, SHH activated, subtype 3", "Medulloblastoma, SHH activated, subtype 4", "Medulloblastoma, SHH activated, IDH mutation", "Medulloblastoma, WNT activation", "Melanocytic tumor", "Melanoma metastasis", "Malignant melanotic nerve sheath tumor", "Meningioma, benign", "Meningioma, subtype intermediate", "Meningioma, malignant subtype", "Meningioma, clear cell subtype, SMARCE1 alteration", "Atypical malignant peripheral nerve sheath tumor", "Malignant peripheral nerve sheath tumor", "Myxoid glioneuronal tumor, PDGFRA mutation", "Neuroblastoma, MYCN type", "Neuroblastoma, TMM negative", "Neuroblastoma, ALT / TERT TMM-positive", "neuroepithelial tumor, MN1:CXXC5 fusion", "neuroepithelial tumor with PATZ1 fusion", "neuroepithelial tumor,"PLAGL1 fusion", "plexiform neurofibroma", "oligodendroglioma, IDH mutation and 1p / 19q co-deletion", "oligosarcoma, IDH mutant", "olfactory neuroblastoma, IDH wild-type", "pilocytic astrocytoma, hemispheric", "pilocytic astrocytoma, infratentorial", "pilocytic astrocytoma, infratentorial, FGFR1 altered subtype", "pilocytic astrocytoma, midline", "pinealoblastoma, MYC / FOXR2 activation", "pinealoblastoma, altered miRNA processing 1 subtype, subtype A", "pinealoblastoma "Pineoblastoma, miRNA processing altered 1 subtype, subtype B", "Pineoblastoma, miRNA processing altered 2 subtype", "Diffuse childhood high-grade glioma, H3 and IDH wild-type, subtype A", "Diffuse childhood high-grade glioma, H3 and IDH wild-type, subtype B", "Diffuse childhood high-grade glioma, MYCN subtype", "Diffuse childhood high-grade glioma, RTK subtype", "Papillary glioneuronal tumor, PRKDC fusion", "Pineocytoma", "Pineoblastoma, RB1 altered subtype", "Pituitary adenoma , ACTH-secreting subtype", "Pituitary adenoma, gonadotropin-secreting subtype", "Pituitary adenoma, prolactin-secreting subtype", "Pituitary adenoma, STH-secreting subtype, dense granular A subtype", "Pituitary adenoma, STH-secreting subtype, dense granular B subtype", "Pituitary adenoma, STH-secreting subtype, sparse granular subtype", "Pituitary adenoma, TSH-secreting subtype", "Pituitary cell tumor, sellar granular cell tumor, spindle oncocytoma", "CNS plasmacytoma", "Pleomorphic low-grade neuroepithelial tumor of the young", "Moderately differentiated pineal parenchymal tumor, KBTBD4 alteration, "Subtype A", "Moderately differentiated pineal parenchymal tumor, KBTBD4 altered, subtype B", "Pinatal papillary tumor, subtype A", "Pinatal papillary tumor, subtype B", "Pleomorphic xanthoastrocytoma", "Retinoblastoma", "Retinoblastoma, MYCN altered subtype", "Rosette-forming glioneuronal tumor", "Rhabdomyosarcoma, MYOD1 mutant", "Neurilemma", "Subependymal giant cell astrocytoma, TSC1 / TSC2 altered", "Solitary fibrous tumor", and "Undifferentiated sinonasal carcinoma, IDH2 mutant";, The training method of the classifier model includes: Obtaining a pre-constructed training set, wherein the training set includes DNA methylation sample data of each preset type; and the DNA methylation sample data carries a label corresponding to the preset type; Inputting the preset type of DNA methylation sample data into a pre-constructed classifier initial model, and iteratively training the classifier initial model using a stochastic gradient descent algorithm until a preset loss function converges, thereby obtaining the classifier model; The training method further comprises: Acquire the pre-collected DNA methylation raw sample data, and assign a label to each of the DNA methylation raw sample data; wherein each of the DNA methylation raw sample data includes a sample size, DNA methylation raw data, and DNA unmethylation raw data; Preprocessing the DNA methylation raw sample data according to a preset preprocessing rule to obtain the DNA methylation sample data; Wherein, if the DNA methylation data is DNA methylation data of a DNA methylation chip, the preprocessing rules at least include: fluorescence correction rules, signal background correction rules, probe filtering rules and signal intensity standardization rules; The fluorescence correction rule is: respectively calculating the average signal intensity values ​​of the DNA methylation probe set and the DNA unmethylation probe set; respectively calculating the DNA methylation normalization factor and the DNA unmethylation normalization factor based on the average signal intensity values; and using the DNA methylation normalization factor and the DNA unmethylation normalization factor to normalize the DNA methylation raw data and the DNA unmethylation raw data in the DNA methylation raw sample data; The signal background correction rule is: the signal intensity value of the DNA methylation probe and the signal intensity value of the DNA unmethylation probe are respectively subtracted from the DNA methylation and DNA unmethylation signal intensity values ​​of the 5% percentile probes in the negative control probe set arranged from small to large signal intensity, and the minimum signal value is set to 0; The probe filtering rules are as follows: filter the probes located on the sex chromosomes; filter the probes with SNP sites within 5 bp of the target CpG site; filter the hybridization non-unique probes; if the DNA methylation data is derived from the GPL13534 chip, the GPL33022 chip and the GPL21145 chip, filter out the probes that are not included in the GPL13534, GPL33022 chip and GPL21145 chip at the same time; The signal intensity standardization rule is: make the standard deviation of the signal intensity of all detection probes of each preset type of sample be 1 and the average value be 0.

2. The classification device according to claim 1, characterized in that: The DNA methylation data includes DNA methylation data of a DNA methylation chip or second-generation sequencing data; The DNA methylation chip includes at least one of the GPL13534, GPL33022 and GPL21145 chips.

3. The classification device according to claim 1, characterized in that: The training method further comprises: Acquire a pre-constructed validation set, wherein the validation set includes DNA methylation sample data of each preset type; and the DNA methylation sample data carries a label corresponding to the preset type; Inputting the DNA methylation sample data included in the validation set into the trained classifier model to obtain a classification result of the classifier model; Counting the classification results, calculating an evaluation index based on the statistical classification results, and evaluating the trained classifier model based on the evaluation index; The evaluation index includes at least one of the following indicators: Comprehensive accuracy: the ratio of the sample size of the DNA methylation sample data that is correctly classified to the total sample size; Recall rate of each preset type: for each preset type, the proportion of the sample volume of the DNA methylation sample data that is actually correctly classified to the required sample volume is counted; The accuracy rate of each preset type: for each preset type, the ratio of the sample volume of the DNA methylation sample data that is actually correctly classified to the total volume of all the DNA methylation sample data classified into the preset type is counted; AUC value: The DNA methylation sample data that is correctly classified is configured as a positive label, and the DNA methylation sample data that is incorrectly classified is configured as a negative label to calculate the AUC value.

4. The classification device according to claim 1, characterized in that: The step of determining the type to which the DNA methylation data of the sample to be tested belongs based on the classification result comprises: The preset type corresponding to the maximum probability in the classification results is selected as the type to which the DNA methylation data of the sample to be tested belongs.

5. The classification device according to claim 1, characterized in that: The step of inputting the DNA methylation data of the sample to be tested into a pre-trained classifier model comprises: The DNA methylation data of the sample to be tested is preprocessed according to the preprocessing rules, and the preprocessed DNA methylation data is input into a pre-trained classifier model.

6. Use of the classification device according to any one of claims 1 to 5 in the preparation of a product for diagnosing central nervous system tumor types.

Images