Multimodal scoring system and application for GBM prognosis and DCV treatment assessment

Abstract

This application relates to the interdisciplinary field of biomedicine and medical artificial intelligence, disclosing a multimodal scoring system and its application for assessing GBM prognosis and DCV treatment. The invention provides a biomarker combination including EFEMP1, PMP22, HRH1, and PTX3, and also discloses a transcriptome scoring model constructed based on this biomarker combination, a radiomics and deep learning non-invasive prediction model based on preoperative MRI, and an immunohistochemical scoring model. Each model has a cutoff value to distinguish between MMR suppression and activation states. The invention also provides the application of the above models in the preparation of kits or computer-aided diagnostic systems for predicting the response of glioblastoma patients to dendritic cell vaccine therapy. Through this invention, it is possible to non-invasively and accurately screen for patients who are sensitive to dendritic cell vaccine therapy, achieving personalized stratified treatment.

Description

Technical Field

[0001] This application belongs to the interdisciplinary field of biomedicine and medical artificial intelligence, and in particular relates to a multimodal scoring system and its application for evaluating the prognosis of glioblastoma and the response to dendritic cell vaccine treatment. Background Technology

[0002] Glioblastoma (GBM) is the most common and most malignant primary tumor of the adult central nervous system, with extremely high mortality and disability rates. Despite current comprehensive treatment options including surgery, radiotherapy, and chemotherapy, the median survival for patients remains only about 20 months. In recent years, immunotherapy, represented by dendritic cell vaccines (DCV), has shown promising application prospects; however, its clinical response rate exhibits significant individual variability, and there is currently a severe lack of reliable biomarkers in clinical practice capable of accurately identifying the population with the greatest potential for benefit.

[0003] In various solid tumors, mismatch repair deficiency (dMMR) or high microsatellite instability (MSI-H) has been proven to be effective biomarkers for predicting the efficacy of immune checkpoint inhibitors. However, in GBM, the mutation rate of traditional MMR genes (such as MLH1, MSH2, MSH6, and PMS2) is extremely low, usually only seen in sporadic cases or associated with Lynch syndrome. Therefore, traditional binary assessment methods based on gene mutations or the deletion of four proteins are difficult to accurately reflect the true functional status of MMR pathways and their immune complexity in the GBM tumor microenvironment. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a combination of biomarkers for assessing the mismatch repair (MMR) functional status of glioblastoma (GBM), and a multimodal quantitative evaluation system based on this biomarker combination, encompassing transcriptomics, radiology, and pathology. This system does not rely on traditional gene mutation detection but precisely quantifies the MMR functional status of GBM patients at the functional transcriptional and multimodal phenotypic levels. It can independently predict prognosis and highly accurately identify the population that benefits from dendritic cell vaccine (DCV) therapy.

[0005] In a first aspect, this application provides a combination of biomarkers for assessing the MMR function and immune microenvironment status of glioblastoma, the combination of biomarkers including EFEMP1, PMP22, HRH1 and PTX3.

[0006] Secondly, this application provides a kit for evaluating the MMR function and immune microenvironment status of glioblastoma, the kit comprising reagents for detecting the expression levels of the above-mentioned combination of biomarkers.

[0007] Thirdly, this application provides a transcriptome scoring model, which adopts the following technical solution: A transcriptome scoring model, wherein the model is based on the mRNA expression level of the above-mentioned biomarker combination, and the MMR score is calculated by principal component analysis; the MMR score is calculated by the formula: MMR score = PC1 + PC2, where PC1 and PC2 are the first two principal components extracted after principal component analysis of the expression levels of four genes EFEMP1, PMP22, HRH1 and PTX3.

[0008] Optionally, when the MMRscore is greater than -0.807, it indicates a mismatch repair suppression state; when the MMRscore is less than or equal to -0.807, it indicates a mismatch repair activation state.

[0009] Fourthly, this application provides a non-invasive prediction model based on radiomics and deep learning, employing the following technical solution: A non-invasive prediction model based on radiomics and deep learning is proposed. The model is based on the patient's preoperative T1-weighted enhanced magnetic resonance imaging (MRI) images and constructs MMRscore_Radio by fusing radiomics features and deep learning features. The radiomics features include first-order statistics, shape features, texture features, and wavelet features. The deep learning features are based on a pre-trained ResNet-50 network and employ a slice-by-slice strategy to extract high-level semantic features and perform global average pooling.

[0010] Optionally, the MMRscore_Radio is calculated as follows: the fused features are subjected to Spearman correlation analysis and the maximum correlation minimum redundancy algorithm for dimensionality reduction, and the top 10 features are retained and input into the LASSO classifier to obtain the predicted score.

[0011] Optionally, when the MMRscore_Radio score is greater than 0.785, it indicates a mismatch repair suppression state; when the MMRscore_Radio score is less than or equal to 0.785, it indicates a mismatch repair activation state.

[0012] Fifthly, this application provides an immunohistochemical scoring model, wherein the MMRscore_IHC is obtained by performing principal component analysis and summing the H-scores of the four proteins EFEMP1, PMP22, HRH1 and PTX3 in claim 1 through immunohistochemical staining.

[0013] Sixthly, this application provides the application of the above-mentioned transcriptome scoring model, the above-mentioned non-invasive prediction model, or the above-mentioned immunohistochemical scoring model in the preparation of a kit or computer-aided diagnostic system for predicting the response of GBM patients to DCV treatment.

[0014] In summary, this application includes at least one of the following beneficial technical effects: 1. Breaking through the binary limitation: For the first time, the two functional states of "MMR activation" and "MMR inhibition" that are not mutation-dependent were systematically characterized in GBM, solving the problem of extremely low positive rate in traditional mutation detection; 2. Cross-modal clinical applicability: A highly consistent "three-in-one" assessment framework integrating transcriptomics, radiology, and pathology was constructed; in particular, the introduction of preoperative MRI imaging models enabled fully non-invasive and rapid preoperative intervention decisions. 3. Precision-guided immunotherapy: It reveals that although patients with high MMR scores have a poor prognosis, their tumor microenvironment is in a state of "active but suppressed antigen presentation". Clinical trials have confirmed that this group can obtain significant survival benefits from DCV treatment, which has extremely high clinical translational value. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the overall process of building and validating the multimodal MMRscore system of the present invention. Figure 2 This is a network diagram showing the correlation analysis (Mantel test) between the expression levels of the core hub gene combination (including EFEMP1, PMP22, HRH1, and PTX3) and classical mismatch repair (MMR) genes and the abundance of various immune cell infiltration in the tumor microenvironment in Example 2. Figure 3 The plot shows the receiver operating characteristic (ROC) curves of the MMRscore_Radio model based on preoperative MRI in Example 3 on different validation sets. Figure 4 This is a graph showing the results of the clinical cohort that received DCV treatment in Example 5. Detailed Implementation

[0016] The following embodiments are provided to better understand the present invention and are not limited to the preferred embodiments described. They do not constitute a limitation on the content and scope of protection of the present invention. Any product that is the same as or similar to the present invention, derived by any person under the guidance of the present invention or by combining the features of the present invention with other prior art, falls within the protection scope of the present invention.

[0017] When numerical ranges are given in the embodiments, it should be understood that, unless otherwise stated in the present invention, both endpoints of each numerical range and any value between the two endpoints may be selected. Unless otherwise defined, all technical and scientific terms used in this invention, as well as the prior art known to those skilled in the art and the description of this invention, may be implemented using any prior art methods, devices, and materials similar to or equivalent to those described, used, or made of materials in the embodiments of this invention.

[0018] To explain the technical content, objectives, and effects of the technical solution in detail, the following description is provided in conjunction with specific embodiments. The overall flowchart for the construction and verification of the multimodal MMRscore system is shown below. Figure 1 As shown. Unless otherwise specified, the materials and equipment used in the various embodiments of this application are all commercially available products in the art.

[0019] Example 1: Construction and application of a quantitative transcriptome scoring system (MMRscore) Surgical resection samples were obtained from GBM patients, and total RNA was extracted and subjected to transcriptome sequencing (RNA-seq). After standardization of the sequencing data, the expression levels of mRNAs of four genes, EFEMP1, PMP22, HRH1, and PTX3, were extracted.

[0020] Total RNA was extracted from the patient's tumor tissue and sequenced to obtain the gene expression matrix. The first two principal components were extracted and summed using the principal component analysis (PCA) dimensionality reduction algorithm module. The formula is: MMRscore = PC1 + PC2.

[0021] The optimal cutoff value is determined to be -0.807 based on the "surv_cutpoint" function. MMRscore > -0.807 is defined as a high score indicating "MMR suppression", and MMRscore ≤ -0.807 is defined as a low score indicating "MMR activation".

[0022] Example 2: Validation of the mechanism by which biomarker combinations reflect the state of the immune microenvironment Mantel test network correlation analysis was performed on the four extracted core hub gene combinations (covering EFEMP1, PMP22, HRH1, and PTX3), and the results are as follows: Figure 2 As shown.

[0023] The results showed that the biomarker combination was significantly correlated with the expression of several classic MMR-related genes (such as MSH2, MSH6, and PCNA), confirming its potential as a molecular surrogate indicator for evaluating MMR functional status. More importantly, the hubgene combination was highly positively correlated with the infiltration abundance of cell populations closely related to antigen presentation and immune activation (such as activated dendritic cells, macrophages, and effector T cells) (Mantel's r ≥ 0.1, p ≤ 0.01). This result indicates that the biomarker combination can effectively map the state of the tumor's immune microenvironment, providing solid underlying biological evidence for accurately screening and predicting the response of glioblastoma patients to dendritic cell vaccine (DCV) immunotherapy.

[0024] Example 3: Multimodal imaging prediction model based on preoperative MRI (MMRscore_Radio) A method for non-invasive prediction of MMR status based on preoperative T1-weighted enhanced (T1+C) MRI images includes the following steps: Image standardization preprocessing: Acquire preoperative T1-weighted enhanced (T1+C) MRI images of the patient, apply N4 bias field correction and resample uniform voxels; Region of Interest (ROI) segmentation: The physician manually delineates the tumor ROI and it is then reviewed by a senior expert. Extracting traditional radiomics features: using PyRadiomics to extract first-order statistics, shape, texture, and wavelet features; Deep learning semantic features based on ResNet-50 network: Based on the pre-trained ResNet-50 network, a slice-wise strategy is used to extract high-level semantic features and perform global averaging; Dimensionality reduction using minimum redundancy maximum correlation (mRMR) algorithm: By fusing the above features, the top 10 features are retained after dimensionality reduction using Spearman correlation and mRMR algorithms. Output predicted scores using the LASSO regression classifier: Input the LASSO classifier to calculate the final predicted score.

[0025] The results are as follows Figure 3 As shown, in the training set of this embodiment, the optimal cutoff value is determined to be 0.785.

[0026] Example 4: Construction of the Pathological Grade Scoring System (MMRscore_IHC) Immunohistochemical staining of EFEMP1, PMP22, HRH1, and PTX3 was performed on routine paraffin-embedded tissue sections using specific antibodies. Pathologists independently evaluated the samples under a microscope in a blinded manner, calculating the H-score for each protein. These four H-scores were then input into a principal component analysis (PCA) module to extract and sum the principal component scores to obtain the MMRscore_IHC score, which is used as an alternative diagnostic tool in scenarios lacking sequencing and imaging analysis capabilities.

[0027] Example 5: Validation of the application of a scoring system in evaluating the efficacy of dendritic cell vaccine (DCV) In a randomized controlled clinical trial (e.g., using the HS-GBM-DCV cohort at Huashan Hospital as the validation set), GBM patients undergoing total tumor resection were randomly assigned to the DCV treatment group and the placebo group. Patient scores were calculated using the MMRscore_Radio model from Example 3. Results are as follows... Figure 4 As shown, survival analysis revealed that patients in the high-scoring group (score > 0.785) had a significantly longer overall survival (OS) after DCV treatment compared to the placebo group (p = 0.035); while patients in the low-scoring group did not experience a statistically significant survival benefit from DCV treatment. Similarly, using the MMRscore_IHC score from Example 4 yielded highly consistent efficacy prediction results (p = 0.041). This confirms the practical utility of this system in the precise clinical screening of vaccine recipients.

[0028] The specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. Those skilled in the art can make modifications to these embodiments without contributing any inventive step after reading this specification, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

Claims

1. A combination of biomarkers for assessing MMR function and immune microenvironment status in glioblastoma, characterized in that, The biomarker combination includes EFEMP1, PMP22, HRH1, and PTX3.

2. A kit for assessing the MMR function and immune microenvironment status of glioblastoma, characterized in that, The kit includes reagents for detecting the expression levels of the biomarker combination of claim 1.

3. A transcriptome scoring model, characterized in that, The model is based on the mRNA expression level of the biomarker combination described in claim 1, and calculates the MMR score through principal component analysis; the formula for calculating the MMR score is: MMR score = PC1 + PC2, where PC1 and PC2 are the first two principal components extracted after principal component analysis of the expression levels of the four genes EFEMP1, PMP22, HRH1 and PTX3.

4. The transcriptome scoring model as described in claim 3, characterized in that, When the MMR score is greater than -0.807, it indicates a mismatch repair suppression state; when the MMR score is less than or equal to -0.807, it indicates a mismatch repair activation state.

5. A non-invasive prediction model based on radiomics and deep learning, characterized in that, The model is based on the patient's preoperative T1-weighted enhanced magnetic resonance imaging images and constructs MMRscore_Radio by fusing radiomics features and deep learning features. The radiomics features include first-order statistics, shape features, texture features, and wavelet features. The deep learning features are based on a pre-trained ResNet-50 network and use a slice-by-slice strategy to extract high-level semantic features and perform global average pooling.

6. The non-invasive prediction model as described in claim 5, characterized in that, The MMRscore_Radio is calculated as follows: the fused features are subjected to Spearman correlation analysis and the maximum correlation minimum redundancy algorithm for dimensionality reduction, and the top 10 features are retained and input into the LASSO classifier to obtain the predicted score.

7. The non-invasive prediction model as described in claim 6, characterized in that, When the MMRscore_Radio score is greater than 0.785, it indicates a mismatch repair suppression state; when the MMRscore_Radio score is less than or equal to 0.785, it indicates a mismatch repair activation state.

8. An immunohistochemical scoring model, characterized in that, In the model, the MMRscore_IHC is obtained by performing principal component analysis and summing the H-scores of the four proteins EFEMP1, PMP22, HRH1 and PTX3 in claim 1 through immunohistochemical staining.

9. The application of the transcriptome scoring model as described in claim 3, the non-invasive prediction model as described in claim 5, or the immunohistochemical scoring model as described in claim 8 in the preparation of kits or computer-aided diagnostic systems for predicting the response of GBM patients to DCV treatment.

Images