A method for predicting TP53 mutation typing and its application

Through multi-platform analysis and machine learning technology, we can identify the immunosuppressive characteristics of COCA subtypes, especially the COCA3 subtype, in patients with TP53 mutation, establish a neural network model, predict its sensitivity to ICB treatment, and solve the problem of selecting drug-resistant patients The problem provides an effective treatment plan, emphasizes the concept of tumor suppression, and has important clinical guiding significance.

CN115273975BActive Publication Date: 2024-11-22THE FIRST HOSPITAL OF CHINA MEDICIAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210936165.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-08-05
Publication Date
2024-11-22
Estimated Expiration
2042-08-05

AI Technical Summary

Technical Problem

There is difficulty in selecting TP53 mutation patients who are resistant to immune checkpoint blockade (ICB) therapy. The existing research focuses on the molecular, biological and clinical differences in whether TP53 mutations, heterogeneity and malignant progression patterns of TP53 mutation tumors Limited understanding.

Method used

Through comprehensive multi-platform analysis, five COCA subtypes in patients with TP53 mutation were identified, especially the COCA3 subtype with immunosuppressive microenvironment, and a neural network model was established using machine learning technology to distinguish COCA3 from other subtypes, and predict their counterparts. Sensitivity of ICB treatment.

Benefits of technology

It provides a robust predictive model that can verify its effectiveness in multiple large databases, has important clinical guidance, helps select patients with TP53 mutations who are sensitive to ICB treatment, provides relevant treatment plans, and emphasizes the concept of tumor suppression. It has strong clinical transformation value.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115273975B_ABST
    Figure CN115273975B_ABST
Patent Text Reader

Abstract

The present invention belongs to the field of medicine and health industry, and specifically relates to a method for exploring different subtype characteristics of TP53 mutant tumors, which is a method for implementing precision medicine based on pan-cancer multi-omics variation characteristics. The present invention conducted a comprehensive multi-platform analysis of patients with TP53 mutations and identified 5 COCA subtypes. Each subtype exhibits different genetic and epigenetic characteristics, resulting in different biological functions and clinical outcomes. In particular, COCA3 has an immunosuppressive microenvironment, resulting in poor ICB response. Based on the COCA clustering characteristics, we used machine learning to establish a robust model to distinguish between COCA3 subtypes and other TP53 mutant patients, and verified it through an immunotherapy database. The prediction model in the present invention has the advantages of being simple and easy to operate, low cost, and is supported by multiple large databases. The evaluation efficiency is relatively stable, and it has a strong clinical translation value in guiding treatment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of medicine and health industry, and specifically relates to a method for exploring the different subtype characteristics of TP53 mutant tumors, which is a method for implementing precision medicine based on pan-cancer multi-omics variation characteristics. Background Art

[0002] The TP53 tumor suppressor gene is mutated in a variety of human cancers, with mutation rates ranging from less than 5% to more than 90%. As one of the most important molecular indicators, TP53 mutation is always associated with genomic instability, cell cycle dysregulation, and poor clinical prognosis. Previous studies have mainly been aimed at clarifying the molecular, biological, and clinical differences between tumors with and without TP53 mutations. However, there is also great heterogeneity between tumors with TP53 mutations. Our understanding of the progression pattern of malignant tumors under TP53 mutations is still limited.

[0003] In recent years, immune checkpoint blockade (ICB) therapy has shown unprecedented durable clinical benefit rates in patients with various cancers. Although patients with TP53 mutations generally have a poor prognosis, most of them have been found to be sensitive to ICB treatment. However, this benefit is not widely applied to all patients with TP53 mutations. How to select patients with TP53 mutations who are resistant to ICB remains challenging. Summary of the invention

[0004] In response to the above problems, the present invention conducted a comprehensive multi-platform analysis of patients with TP53 mutations for the first time and identified 5 COCA subtypes. Each subtype exhibits different genetic and epigenetic characteristics, leading to different biological functions and clinical outcomes. In particular, COCA3 has an immunosuppressive microenvironment, resulting in poor ICB response. Based on the COCA clustering characteristics, we used machine learning to establish a robust model to distinguish between COCA3 subtypes and other patients with TP53 mutations, and verified it through the immunotherapy database. In summary, our analysis can fully provide evidence for pre-existing hypotheses and provide relevant treatment options for patients with TP53 mutations.

[0005] In order to achieve the above object, the present invention provides the following technical solutions.

[0006] The present invention provides a method for establishing a prediction model of tumor TP53 gene mutation COCA3 subtype sensitivity to immunotherapy based on machine learning technology, characterized in that the method comprises the following steps:

[0007] Step 1: Based on big data, cluster analysis was performed on the mutation, SCNA and methylation data of the TP53 gene to identify five subtypes with different genomic and epigenomic characteristics, which were named: COCA1, COCA2, COCA3, COCA 4, and COCA 5;

[0008] Step 2: A neural network learning algorithm based on machine learning technology was used to train the training set and the characteristic genes that accurately separated COCA3 from other subtypes to establish a prediction model for the sensitivity of tumor TP53 gene mutation COCA 3 subtype to immunotherapy;

[0009] Step 3: After the model training is completed, the immunotherapy sensitivity test is completed on the validation set according to the above steps to obtain the test results.

[0010] Furthermore, step 1 of the method for tumor TP53 gene mutation typing comprises the following steps:

[0011] 1) Data processing: The TP53 mutation frequency in 9104 samples from the Cancer Genome Atlas project and in different tumor types was analyzed, and finally TP53 mutation-related data for 2773 samples including 18 types of cancer were obtained;

[0012] 2) Data integration and analysis: Unsupervised clustering analysis was performed based on the mutation, SCNA and methylation data of the TP53 gene, and each platform generated 6, 13 and 6 subtypes, respectively. Clustering and re-clustering analysis was performed on all TP53 mutation samples based on the classification results of mutation, SCNA and methylation, and 5 subtypes with different genomic and epigenomic characteristics were identified, which were named: COCA1, COCA 2, COCA 3, COCA 4, and COCA 5.

[0013] Furthermore, in step 2 of the method, the characteristic genes for accurately separating COCA3 from the other four subtypes are screened: lasso regression with the best alpha value is applied to the TCGA mRNA-seq cohort to screen out the characteristic genes for accurately separating COCA1, 2, 4, 5 and COCA3.

[0014] Furthermore, the optimal alpha value is 0.0015.

[0015] Furthermore, the step of neural network learning in step 2 of the method includes:

[0016] 1) We first used lasso regression with the best alpha value (0.0015) on the TCGA mRNA-seq cohort and selected 489 signature genes that could separate COCA3 from other classifications through the inner layer of a 10,000-fold five-fold cross-validation process;

[0017] 2) The neural network was built based on keras with random starting parameters. The input of the first fully connected layer was 489 gene expressions and the output was 500 neurons; the input of layer2 was 500 and the output was 1500; the input of layer3 was 1500 and the output was 1700; the input of layer4 was 1700 and the output was two categories, namely COCA3 and other categories;

[0018] 3) The important parameters of the model setting are as follows: batchsize=200, epochs=50, optimizer=adam, learning rate=0.001.

[0019] Furthermore, the COCA 3 subtype has an immunosuppressive microenvironment and an immunosuppressive phenotype, and is a robust subtype with a poor prognosis.

[0020] Compared with the prior art, the present invention has the following beneficial effects.

[0021] 1) Existing technologies all study the differences in characteristics between TP53 mutant and non-mutated tumors. The present invention systematically explores the molecular characteristics of different subtypes of TP53 mutant tumors and emphasizes the concept of tumor suppression.

[0022] 2) The neural network model for predicting the third category provided by the present invention has a strong and stable predictive ability and has been verified in multiple large sample databases. It can be used as a reliable indicator for predicting prognosis in clinical practice and has important clinical guiding significance.

[0023] 3) The prediction model in the present invention has the advantages of being simple, easy to operate, and low-cost, and is supported by multiple large databases. The evaluation efficiency is relatively stable and has strong clinical translation value in guiding treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A total of 1,773 patients with TP53 mutations in 18 types of tumors were screened for follow-up studies.

[0025] Figure 2 Multi-omics clustering study of patients with TP53 mutations.

[0026] Figure 3 Functional exploration and prognostic study of five TP53 subtypes.

[0027] Figure 4 The third class of TP53 mutations possesses an M2-enriched immunosuppressive microenvironment.

[0028] Figure 5 Multiple databases validated the insensitivity of three types of immunotherapy. DETAILED DESCRIPTION

[0029] The following examples will help to understand the present invention, but these examples are only for illustrating the present invention, and the present invention is not limited to these contents. The operating methods in the examples are all conventional operating methods in the art.

[0030] Example 1 Data processing of patients with TP53 mutations.

[0031] To systematically explore the characteristics of TP53 mutant tumors, we designed an overall flow chart. We retrospectively analyzed the molecular and clinical information of 9104 patients from the Cancer Genome Atlas (TCGA) project. We first analyzed the mutation frequency of TP53 in tumors and found that approximately 35.9% of patients had TP53 mutations. Among tumor types, the frequency of TP53 mutations ranged from 0% (uveal melanoma) to 91.2% (uterine carcinosarcoma) ( Figure 1 A). Through further screening conditions (TP53 mutation frequency greater than 10% and sample number greater than 50), we screened a total of 19 cancer types that met the requirements. Since only 12 ovarian cancer (OV) patients had complete data, they were excluded. Finally, our study included a total of 2773 patients with TP53 mutations in 18 cancer types ( Figure 1 BE).

[0032] Example 2 Omics integrated analysis of TP53 mutant tumors.

[0033] The present invention explored the molecular alteration patterns in TP53 mutant tumors and performed unsupervised cluster analysis (mutation, copy number alteration, and DNA methylation) based on genomic and epigenomic data. Based on mutation, SCNA, and methylation data, each platform generated 6, 13, and 6 subtypes, respectively ( Figure 2 AC). To explore the tumors more comprehensively, we performed cluster-recluster analysis (COCA) on all samples with TP53 mutations based on the classification results of mutation, SCNA, and methylation. The comprehensive analysis identified five subtypes with different genomic and epigenomic characteristics ( Figure 2 D). We found that tumors were divided into COCA subtypes independent of histopathology in patients with TP53 mutations, suggesting that multi-omic clustering formed a novel classification signature ( Figure 2 E).

[0034] Example 3 Biological functions and clinical values ​​of COCA subtypes.

[0035] To explore the biological characteristics of COCA subtypes, we first identified the core genes of each subtype based on weighted gene co-expression network analysis (WGCNA). After removing genes with low standard deviations (<2.0), we obtained a sum of 13 modules containing 6437 genes. We then explored the relationship between modules and COCA subtypes. We selected the most relevant modules as the core modules of each COCA subtype, respectively (P<0.05). Based on the criteria of (MM) ≥0.3 and (GS) ≥0.2, we filtered the core genes that best represented the characteristics of the five subtypes. We then performed GO analysis based on hub genes using Cytoscape and established characteristic biological processes of COCA subtypes (P <0.05) ( Figure 3 A).

[0036] COCA1 is characterized by enrichment of the ERBB2 signaling pathway, regulation of keratinocyte differentiation, and development of lung morphogenesis. COCA2 is enriched in negative regulation of chondrocyte differentiation and positive regulation of proliferation. COCA3 is highly associated with immune response. COCA4 is characterized by inactivated MAPKK activity. COCA5 is associated with negative regulation of lipid metabolism and macrophage proliferation. Thus, we determined the biological phenotypes of COCA isoforms.

[0037] We next evaluated the clinical value of the five clusters. First, we performed survival analysis of TP53-mutated tumors classified by the five COCA subtypes and found the prognostic significance of different subtypes in overall survival (OS) and disease-specific survival (DSS). We then explored the prognostic role of the COCA classification in each cancer type. Among cancers with prognostic significance, we found that the COCA3 subtype had consistent prognostic significance, that is, COCA3 was always associated with a worse prognosis ( Figure 3 In addition, Cox regression analysis also showed that COCA3 had a greater survival disadvantage in LGG, LUSC, and STAD than in the other four subtypes ( Figure 3 H). Taken together, these results highlight the potential clinical utility of this COCA classification system, with COCA3 being a robust subtype with a poorer prognosis.

[0038] Example 4 COCA3 has an immunosuppressive phenotype.

[0039] Considering the poor prognosis of COCA3, we have fully studied the dysregulated immune response as a biological feature of COCA3. Based on the concept of the seven-step cancer immunity cycle (TIP), we evaluated and compared the anti-tumor immune status between COCA3 and the other four COCAs. We found an interesting phenomenon that COCA3 patients have an advantage in activating anti-tumor immunity (the first three steps: cancer antigen release, cancer antigen presentation, priming, and activation). However, in the process from immune cell trafficking to infiltration, COCA3 showed a disadvantage compared with the other four COCA subtypes (Step4 and 5), and then T cell recognition of cancer cells was weakened (Step6), resulting in defects in killing cancer cells (Step7)) ( Figure 4 A). Further GSEA analysis showed that COCA3 was negatively correlated with leukocyte degranulation, indicating its abnormal immune response ( Figure 4 B). We also summarized a total of 75 immune checkpoint genes and scored inhibitory or stimulatory regulatory points based on ssGSEA. We found that the inhibitory checkpoint score was significantly upregulated in COCA3 ( Figure 4 C). Meanwhile, after comparing the stimulation and inhibition scores of each patient, we named the patients as stimulation or inhibition patients and found that COCA3 was significantly enriched in inhibition patients ( Figure 4 D). All these results confirmed that COCA3 was in an immunosuppressive state due to abnormal immune microenvironment and T cell recognition dysfunction, which weakened the anti-tumor effect.

[0040] Based on the different contents of immune cells, we hope to identify the characteristic immune cells that cause the loss of T cell killing ability in COCA3. We compared the infiltration status of immune cells in COCA3 and other COCA subtypes by quantiseq and xcell algorithms. The results showed that B cells, M2 macrophages and CD8+T cells were increased in COCA3 at the same time ( Figure 4 E). Further evaluation of the cell composition showed that the content of M2 macrophages was higher than that of the other two cell types in patients with TP53 mutations, indicating their key role in the immune response ( Figure 4 F). We also found that the content of M2-M1 was increased in COCA3, further shaping the immunosuppressive microenvironment ( Figure 4 GH). In conclusion, we conclude that despite CD8+ T cell infiltration, the COCA3 subtype confers a dynamic polarization of M1 to M2 macrophages, which in part contributes to the reduced anti-tumor capacity.

[0041] Example 5 A classification method for predicting immunotherapy response based on COCA characteristics.

[0042] Since the immune phenotypes among COCAs were different, we next sought to investigate their impact on immunotherapy. We first used the TIDE algorithm to predict clinical response to immunotherapy. We compared the response of COCA3 with the other four clusters to investigate their sensitivity to ICB treatment. As expected, COCA3 was less likely to respond to ICB compared with the other four clusters (P < 0.05) ( Figure 5 A).

[0043] In recent years, artificial intelligence (AI) has been used in an endless stream in the treatment of patients and will also be used to help doctors diagnose and treat various diseases. Considering that RNAseq databases are widely used in clinical and scientific research, we finally established a machine learning-based COCA3 classification method using RNAseq. Before learning, we first applied lasso regression with the best alpha value (0.0015) on the TCGA mRNA-seq cohort and selected 489 characteristic genes that can accurately separate COCA1.2.4.5 and COCA3 (Table 1). Then, the patients were divided into training and validation sets at a ratio of 80%:20%, and the training set will be further trained based on these characteristic genes using a variety of machine learning algorithms (SVM, LDA, CART, NB, and neural network) including k-nearest neighbor (KNN), support vector machine. We finally chose the neural network as the perfect learning algorithm because it had the highest prediction accuracy (100% in the training cohort and 95.1% in the independent validation cohort) to classify COCA3 and other COCA groups.

[0044] Table 1. 489 predicted third class genes

[0045]

[0046]

[0047] .

[0048] To verify the clinical significance of our machine learning approach, we analyzed the ICGC database for further validation. We first predicted the COCA subtype of each TP53 mutation patient and found that patients in COCA3 consistently had higher M2 and M2-M1 infiltration ( Figure 5 BC). In addition, we calculated the TIDE score for each patient with TP53 mutation and determined that COCA3 patients had significantly increased TIDE scores, indicating a poor benefit from ICB treatment ( Figure 5D). We then projected the Imvigor210 cohort generated from a phase II trial of atezolizumab in advanced or metastatic urothelial carcinoma (mUC). When comparing the immunotherapy responses in the COCA3 and COCA1.2.4.5 groups, we also found that COCA3 had a poorer treatment response ( Figure 5 E). In conclusion, we explored a machine learning-based subtype of patients with TP53 mutations, which had good predictive value for evaluating the efficacy of immunotherapy.

Claims

1. A method for establishing a prediction model for the sensitivity of tumor TP53 gene mutation COCA3 subtype to immunotherapy based on machine learning technology, characterized in that: The steps of the method include: Step 1: Based on big data, cluster analysis was performed on the mutation, SCNA and methylation data of the TP53 gene to identify five subtypes with different genomic and epigenomic characteristics, which were named: COCA1, COCA2, COCA3, COCA4, and COCA5; The TP53 gene mutation typing comprises the following steps: 1) Data processing: The TP53 mutation frequency in 9104 samples from the Cancer Genome Atlas project and in different tumor types was analyzed, and finally TP53 mutation-related data for 2773 samples including 18 types of cancer were obtained; 2) Data integration analysis: Unsupervised clustering analysis was performed based on the mutation, SCNA and methylation data of the TP53 gene, and each platform generated 6, 13 and 6 subtypes, respectively. Clustering and re-clustering analysis was performed on all TP53 mutation samples based on the classification results of mutation, SCNA and methylation, and 5 subtypes with different genomic and epigenomic characteristics were identified, which were named: COCA1, COCA2, COCA3, COCA4, COCA5; Step 2: A neural network learning algorithm based on machine learning technology is used to train the training set and the characteristic genes that accurately separate COCA3 from other subtypes, and a prediction model for the sensitivity of the tumor TP53 gene mutation COCA3 subtype to immunotherapy is established; the COCA3 subtype has an immunosuppressive microenvironment and an immunosuppressive phenotype, and is a robust subtype with poor prognosis; Step 3: After the model training is completed, the immunotherapy sensitivity test is completed on the validation set according to the above steps to obtain the test results.

2. According to claim 1, a method for establishing a prediction model of tumor TP53 gene mutation COCA3 subtype sensitivity to immunotherapy based on machine learning technology, characterized in that: In step 2 of the method, the characteristic genes for accurately separating COCA3 from the other four subtypes are screened: lasso regression with the best alpha value is applied to the TCGA mRNA-seq cohort to screen out the characteristic genes for accurately separating COCA1, 2, 4, 5 and COCA3.

3. The method according to claim 2, characterized in that The optimal alpha value is 0.0015.

4. The method for establishing a prediction model of tumor TP53 gene mutation COCA3 subtype sensitivity to immunotherapy based on machine learning technology according to claim 1, characterized in that: The steps of the neural network learning algorithm in step 2 of the method include: 1) We first used lasso regression with an alpha value of 0.0015 on the TCGA mRNA-seq cohort and selected 489 signature genes that could separate COCA3 from other classifications through the inner layer of a 10,000-fold five-fold cross-validation process; 2) The neural network was built based on keras with random starting parameters. The input of the first fully connected layer was 489 gene expressions and the output was 500 neurons; the input of layer2 was 500 and the output was 1500; the input of layer3 was 1500 and the output was 1700; the input of layer4 was 1700 and the output was two categories, namely COCA3 and other categories; 3) The important parameters of the model setting are as follows: batchsize=200, epochs=50, optimizer=adam, learningrate=0.001.