阅读说明：本技术 一种揭示乳腺癌免疫逃避调控机制的多源数据融合框架 (Multisource data fusion framework for revealing breast cancer immune evasion regulation and control mechanism ) 是由 陈浩文 宁斌 林业雄 屈强 于 2021-01-11 设计创作，主要内容包括：本发明涉及生物信息学中的数据挖掘领域,具体涉及一种揭示乳腺癌免疫逃避调控机制的多源数据融合框架。其发明内容主要包括：(1)收集乳腺癌样本、正常样本相关数据；(2)利用NMF对乳腺癌样本进行聚类,得到样本亚群种类；(3)将乳腺癌样本与从GTEx中获得的正常样本进行比较,找出差异表达的相关基因；(4)设计了一个基于ATAC-SEQ数据的调控分析算法来寻找免疫相关基因；(5)使用五种通用数据库对TF与免疫基因的关系进行验证；(6)分析根据框架得到的免疫基因是否对患者的生存造成影响。本发明提供了一种揭示乳腺癌免疫逃避调控机制的多源数据融合框架,对于研究药物重定位和实现精准医疗具有重要意义。可以有效提升研究过程和研究结果的生物学意义。更重要的是,本发明的单样本规律分析方法可以更深入地探索肿瘤的异质性,对精准医学实践具有重要意义。(The invention relates to the field of data mining in bioinformatics, in particular to a multi-source data fusion framework for revealing an immune escape regulation and control mechanism of breast cancer. The invention mainly comprises the following steps: (1) collecting data related to the breast cancer sample and the normal sample; (2) clustering the breast cancer samples by using NMF to obtain a sample subgroup variety; (3) comparing the breast cancer sample with a normal sample obtained from GTEx to find out related genes with differential expression; (4) designing a regulatory analysis algorithm based on ATAC-SEQ data to search immune related genes; (5) verifying the relation between TF and immune genes by using five general databases; (6) analyzing whether the immune gene obtained according to the framework has influence on the survival of the patient. The invention provides a multi-source data fusion framework for revealing an immune evasion regulation mechanism of breast cancer, and has important significance for researching drug relocation and realizing accurate medical treatment. The biological significance of the research process and the research result can be effectively improved. More importantly, the single-sample rule analysis method can deeply explore the heterogeneity of the tumor and has important significance on precise medical practice.)

1. A multisource data fusion framework for revealing an immune evasion regulation mechanism of breast cancer is characterized by comprising the following implementation steps:

(1) collecting breast cancer sample information, NK cell ligand information, candidate immune gene information, normal breast tissue sample information and ATAC-seq peak signal information about TCGA;

(2) clustering the breast cancer samples to obtain different immune subtypes and the number of each subtype in the breast cancer samples;

(3) comparing the breast cancer subtype with a normal breast sample to find out an immune related gene with differential expression;

(4) a regulation and control analysis algorithm is designed to obtain the regulation and control factor of immune related gene expression.

2. The multi-source data fusion framework for revealing immune evasion regulatory mechanisms of breast cancer according to claim 1, characterized by the implementation steps of:

(1) collecting a breast cancer sample provided by TCGA as a main body of analysis data;

(2) collecting 2171 candidate immune-related and important NK cell ligands;

(3) normal breast tissue samples in GTEx data were used because the number of normal samples in TGCA was too small and there was a bias;

(4) UCSC Xena provides ATAC-seq peak signals for TCGA samples in pan-cancerous fashion.

3. The multi-source data fusion framework for revealing immune evasion regulatory mechanisms of breast cancer according to claim 1, characterized by clustering stages of breast cancer samples:

(1) due to tumor heterogeneity, non-Negative Matrix Factorization (NMF) was used. The rows of the matrix represent the weights of each feature contributing to each cluster, and the columns of the matrix represent the weights of each sample belonging to each cluster;

(2) the estimation of the NMF approximate solution can be seen as an optimization problem in the formula min (| F-WH | + γ R (W, H)), the first component being used to measure the quality of the approximation, the second component using a regularization function to ensure sparsity or smoothness of the matrices W and H;

(3) to identify distinctly different breast cancer sample subpopulations, different r-values were tried, and the best r-value was then selected based on the quality metric of the clustering results.

4. The multi-source data fusion framework of claim 1, wherein the fusion framework is implemented by the steps of:

(1) the expression of immune-related genes of each subtype was compared to normal samples collected from GTEx. In order to eliminate false positive, the candidate pathogenic gene is used as the consensus result of the Edger and the DESeq 2;

(2) spearman correlation coefficient was used to measure the strength of association between the ligand gene itself CNV and mRNA. If the correlation between them is strong, it can be explained that the mRNA change is caused by CNV itself, otherwise it means that there may be other regulatory factors.

5. The multi-source data fusion framework for revealing immune evasion regulatory mechanisms of breast cancer according to claim 1, characterized by the implementation steps of, in the regulatory analysis stage:

(1) UCSC Xena provides 190 ATAC-seq peak signals for TCGA samples;

(2) all peaks located at 20kb of the gene TSS site are considered to contain candidate regulatory regions for Transcription Factor (TF) or Repressor Protein (RP). The number of raw matrices downloaded from UCSC Xena is defined as "log 2([ count +5] PM) -qn" per cell;

(3) analysis rule 1: the closer to the target gene position, the larger the possible regulation relationship between the peak region and the gene is;

(4) analysis rule 2: higher peak scores indicate higher RP or TF activity in this region;

(5) analysis rule 3 applies Spearman correlation coefficients to measure the strength of regulatory relationships between target genes and TF or RP in the peak region. If the Spearman correlation coefficient is above 0.4 or below-0.4, there is at least a moderate relationship between the two;

(6) the relationship between TFs and immune genes was verified using a universal database.

Technical Field

The invention relates to the field of data mining in bioinformatics, in particular to a multi-source data fusion framework for disclosing a breast cancer immune evasion regulation mechanism.

Background

Precision medicine is an emerging cancer prevention and treatment strategy that takes into account the individual variability of each patient's genetic basis. With the help of next generation high throughput sequencing technologies, researchers are becoming more familiar with the details of whole genome mutations, and the overall relationship between different omics data is also becoming more systematic. For precise medicine, it is essential to understand the immune escape mechanisms of tumorigenesis, especially when the heterogeneity of tumors significantly affects the effectiveness of immunotherapy. The subtype of the breast cancer is identified according to the immune related genes, which is helpful for understanding the immune escape routes dominant by different subtypes, so that effective treatment measures are implemented aiming at different subtypes.

Tumor heterogeneity refers to the molecular and cellular differences of a single tumor between different tumor patients (intratumoral heterogeneity), and even the differences between different sites of tumor formation in a single patient (intratumoral heterogeneity). However, researchers only know the iceberg corner of tumor heterogeneity, resulting in lack of targeted precision medicine. Breast cancer also exhibits heterogeneity at the molecular and cellular levels, which inhibits the effectiveness of diagnostic, prognostic, or predictive strategies in routine clinical practice. While previous studies clustered breast cancer samples collected by TCGA, it was still unclear how most of the mechanisms of tumor cells regulate the expression of immune genes to escape killing by immune cells. The number of matched normal samples is too small and normal samples from cancer patients still cannot completely replace true normal tissue samples from non-cancer individuals.

In conclusion, the existing method has defects in the research of tumor immune escape mechanisms, and simultaneously each database has respective defects, so that the research on the cancer cell immune escape regulation and control mechanism of the breast cancer cells under the condition of multi-source data is rarely carried out

Disclosure of Invention

In order to understand the regulatory mechanism of breast cancer immune escape, a multi-source data fusion framework for revealing the breast cancer immune escape regulatory mechanism is provided. The present invention seeks to find immune-related genes differentially expressed in tumor tissues by comparing mRNA data for TCGA and GTEx. To find out the cause of the immune gene expression change, we performed correlation analysis of CNV and mRNA, and analyzed the relation of Transcription Factor (TF) and immune target gene based on ATAC-seq data. Then, the relationship between TFs and the immune gene was verified using the general database.

Further, the construction method of the double-layer gene regulation network comprises the following steps:

the method comprises the following steps: subpopulation identification of TCGA breast cancer samples was performed using NMF clustering algorithm. It is noted that immune-related genes are considered as clustering features, and thus different subgroups may have different immune escape pathways.

Step two: to avoid data bias from normal tissues collected from cancer patients, we compared GTEx normal data to each subset of TCGA breast cancer samples to find differentially expressed immune-related genes.

Step three: we designed a regulatory analysis algorithm based on ATAC-SEQ data to find out the regulatory factors of immune-related gene expression changes.

Step four: we analyzed whether immune-related gene expression would have an impact on patient survival.

Further, the first step specifically includes:

1) comparing the difference between the candidate immune-related genes in the tumor tissue and the normal tissue. In addition to TCGA data, we additionally collected 2171 candidate immune-related genes;

2) the method solves the problems of gene expression quantification, bias of specific research, elimination of batch effect and the like, and for this reason, 511 breast cancer samples and 212 normal samples are downloaded for downstream analysis;

3) tumor samples were grouped using non-Negative Matrix (NMF) factorization. 2171 candidate immune-related genes were clustered to 511 breast cancer samples.

Further, step two we used EdgeR and DESeq2 techniques for comparison. To eliminate false positives, I treated candidate disease-causing genes as a consensus of Edger and DESeq 2. We used spearman correlation to examine the correlation between CNV and mRNA of candidate disease-causing genes.

Further, the third step specifically includes:

1) all peaks at 20kb of the gene TSS site are considered to be candidate regulatory regions containing Transcription Factor (TF) or Repressor Protein (RP)

2) The target gene has a plurality of peaks mapped, so a multi-objective optimization strategy is provided for sequencing all the peaks. We looked at each peak from a different perspective, using distance, score and spearman correlation coefficient, respectively. The spearman correlation coefficient calculation formula is as follows:

3) and (3) verifying the relation between the TF and the immune gene by using a general database.

The invention provides an analysis framework for integrating multi-source data, which can effectively improve the biological significance of a research process and a research result. More importantly, the single-sample rule analysis method can deeply explore the heterogeneity of the tumor and has important significance on precise medical practice. Understanding the heterogeneity of tumors (either intratumoral or intratumoral heterogeneity) is an important basis for precise medicine. This is because different subtypes may use completely different immune escape pathways. Not only may there be no effect, but side effects may also occur if the same treatment is used.

The invention collects immune related genes, combines TCGA case samples and GTEx normal samples, and identifies specific immune genes related to different subgroups of breast cancer. A multi-target standard for evaluating the importance of the peaks near the target gene is designed, and a regulation and analysis algorithm for positioning and regulating TF or RP of immune-related gene expression based on ATAC-SEQ data is provided. Meanwhile, the difference of target gene expression is explained on the level of a single sample, which shows that the framework designed by the invention can be used for precise medical services.

Drawings

FIG. 1: main flow chart of multi-source data fusion framework

FIG. 2: correlation of 190 peaks with PVRL2

FIG. 3: survival results of PVRL2

FIG. 4: survival results of CDH1

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in detail below with reference to experiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The following detailed description of the principles of the invention is provided in connection with the accompanying drawings.

As shown in fig. 1, the multi-source data fusion framework for revealing the immune evasion regulation mechanism of breast cancer provided by the embodiment of the present invention includes the following steps:

1) clustering was performed by non-Negative Matrix Factorization (NMF) of breast cancer sample data and immune gene data provided by TCGA to obtain subpopulations in breast cancer data.

2) The resulting subpopulation data was analyzed with normal sample data obtained from GTEx to obtain gene expression data that was different before the two samples.

3) Different gene expression data obtained before and TCGA peak signals provided by UCSC Xena in a pan-cancer mode are analyzed by a supervision analysis based on ATAC-SEQ, and the data are analyzed to research the mode of reaching the immune escape mechanism of the breast cancer.

The principle of application of each step of the present invention is further described below.

1. Clustering breast cancer samples using NMF

To understand how these tumor cells evade immune cell damage, the present invention collected 2171 candidate immune-related genes. 511 breast cancer samples and 212 normal samples, which have been subjected to uniform alignment, gene expression quantification, study specificity bias and batch effect elimination, are downloaded at the same time, and the data are subjected to downstream analysis.

To study tumor heterogeneity, breast cancer samples were clustered using NMF based on candidate immune-related genes. The mathematical formula for NMF is as follows:

X≈WH

the formula represents an approximation of a matrix X containing n features and p samples. Where all entries in X are non-negative. The estimation of the approximate solution of W and H can be seen as an optimization problem in the following equation:

where the first component is used to measure the approximate quality, i.e. the loss function. To avoid overfitting, the second component uses a regularization function to ensure sparsity or smoothness of the matrices W and H. γ is a parameter for balancing these two components.

2. Analysis of differentially expressed genes between subtypes

Different immune escape pathways are possible due to different tumor subtypes. The invention compares the immune related gene expression of each subtype with a normal sample collected by GTEx. To eliminate false positives, we treated candidate disease-causing genes as consensus results for Edger and DESeq 2.

To avoid NK cell damage, there are two possible immunosuppressive strategies for tumor cells: one is to decrease expression of NK activating ligand, and the other is to increase expression of NK inhibiting ligand. The analyzed NK ligand genes differentially expressed for all subtypes are shown in table 1:

table 1: differentially expressed genes of each subtype

Table 1 the first column lists all differentially expressed NK ligand genes of the 4 subtypes identified by the consensus of Edger and DESeq 2.

The second column lists all NK activator ligands in each cluster, respectively.

The third column lists all these NK activator ligands in each cluster, respectively.

The fourth column lists all these NK inhibitor ligands in each cluster separately, as in this case no significantly low expressed genes were found, the invention is indicated using the '-' symbol.

The fifth column lists all NK inhibitor ligands in each cluster.

It can be seen from the table that CDH1 and PVRL2 occur simultaneously in all clusters, which may mean that they play a very important role in immune escape of breast cancer.

Correlation analysis of CNV with mRNA on CDH1 and PVRL2

Spearman correlation coefficient was used to measure the strength of association between the ligand gene itself CNV and mRNA. If the correlation between them is strong, it can be explained that the mRNA change is caused by CNV itself, otherwise it means that there may be other regulatory factors for the mRNA change.

Spearman correlation is used to measure the strength of a monotonic relationship between two variables. Its calculation and significance test are based on two assumptions: the data for both variables are interval or ratio levels or ordinals, and they are monotonically correlated. The value of the spearman correlation coefficient is [ -1, +1], the closer to +1, the stronger the positive correlation is; the closer to-1, the stronger the negative correlation.

CNV of CDH1 has a medium Spireman correlation coefficient with mRNA (correlation coefficient 0.54, adjusted P value 3.12e-05), but has a weak correlation with PVRL2, so other factors are certain to regulate the gene expression of PVRL 2.

4. ATAC-SEQ-based supervision mechanism analysis

The present invention further explores what factors regulate the expression changes of PVRL 2. As shown in fig. 2, approximately 190 peaks around PVRL2 were found. Using our multi-objective peak selection criteria, three candidate peaks (most positively correlated peak, most negatively correlated peak, and closest gene score peak) were retained for downstream analysis.

The invention matches 5 commonly used databases, namely Jaspar, ENCODE, CHEA, MotifMap and TRANSFAC. The verification results are shown in table 2:

table 2: verification correlation between TF and PVRL2

"√" indicates that the corresponding TF has a regulation relationship with the PVRL2, and "-" indicates that the database does not contain the corresponding regulation relationship between the TF and the PVRL 2.

The same regulatory analysis was also applied for CDH1, with the results shown in table 3:

table 3: verification correlation between TF and CDH1

TF	JASPAR	ENCODE	CHEA	MotifMap	TRANSFAC
						AR	-	-	√	-	-
GTF2L	-	-	-	-	-
						IRF2	-	-	-	-	-
NF1	-	-	-	-	-
						NFATC2	-	-	-	-	-
XBP1	-	-	-	-	-
						YY1	-	√	-	-	-

"√" indicates that the corresponding TF has a regulatory relationship with CDH1, and "-" indicates that the database does not contain the corresponding regulatory relationship between TF and PVRL 2.

Survival analysis of CDH1 and PVRL2

If the CDH1 and PVRL2 obtained by the invention are correct for tumor cells to avoid immune evasion information, their expression should affect the survival of the patient. The present invention therefore depicts the survival analysis results of both CDH1 and PVRL 2. As shown in fig. 3 and 4.

In fig. 4, the blue line indicates low expression of PVRL2, the gray line indicates moderate expression of PVRL2, and the red line indicates high expression of PVRL 2. A P value of 0.009 suggested that PVRL2 had a significant effect on the quality of life of breast cancer patients.

For CdH1, in FIG. 4, the dark blue line indicates low expression, the light blue line indicates moderate expression

A large number of researches prove that the occurrence and development of complex diseases such as tumors generally involve the interaction of various factors such as environment, gene mutation and the like. However, single-level omics data make it difficult to systematically and completely reveal how multiple factors interact. Meanwhile, the single source data set is usually limited by factors such as sample population, sample amount and data type, so that the statistical power is insufficient, and repeated correlation research is difficult. Therefore, the invention provides an analysis framework for integrating multi-source data, and the biological significance of the research process and the research result can be effectively improved. More importantly, the single-sample rule analysis method can deeply explore the heterogeneity of the tumor and has important significance on precise medical practice.

The invention collects immune related genes, combines TCGA case samples and GTEx normal samples, and identifies specific immune genes related to different subgroups of breast cancer. A multi-target criterion for evaluating the importance of peaks near target genes is designed, and a regulation and analysis algorithm for positioning and regulating TF or RP of immune-related gene expression based on ATAC-SEQ data is provided. Differences in target gene expression were explained at the single sample level, indicating that the framework designed by the present invention can be a precise medical service. The present invention uses a statistical method to determine whether the expression level of a target gene in a single sample is higher than that in a normal sample, and then analyzes the specific cause of the expression change of the target gene in a single sample manner.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.