Gene information and bioinformatics analysis
We downloaded gene expression data from TCGA (workflow type: HTSeqCounts) and GTEx (data format: fpkm). It contains 698 tumor samples and 1152 normal samples. Clinical data (1114 cases, data format: BCR XML) was downloaded from TCGA, which included survival time, survival state, age gender and grade. Additionally, the Ensembl database (http://asia.ensembl.org/index.html) was used to transform Ensembl IDs to corresponding gene symbols. Perl language was applied to integrate data and get data matrix, simultaneously, the listwise technology was a great way to deal with missing data. In terms of hypothesis test, we use Kruskal-Wallis test to determine the difference situation, especially clinical significantly relation.
Identification of differentially immune-related expressed genes and immune-gene-pair.
After getting gene transcription data, we used “edgeR” R package (R version:3.6.1) to filter differentially expressed genes between GBM samples and normal samples. ImmPort (http://www.immport.org) is an open database which includes 17 categories, immune-related gene list was downloaded from this web [10]. Our study extracted differentially immune-related genes and their expression level. Afterwards, immune-gene-pair was constructed and it was combined with survival data. By the way, gene-pair meant comparing adjacent genes’ expression levels and obtain results.
KEGG and GO pathways analysis
The Kyoto Encyclopedia of Genes and Genomes (KEGG) is a database which is utilized to explore genes’ biological functions [11]. Gene Ontogy (GO) function analysis includes biological processes (BP), cellular components (CC) and molecular functions (MF), which could let us have a better understanding of biological processes [12]. By using "colorspace", "stringi", "ggplot2", "clusterProfiler" and "enrichplot" R package, we could perform GO functional annotations and KEGG pathway enrichment analysis of differential expressed genes we got before. Results could reveal pathways worthy of attention.
Lasso regression
Lasso regression is a method that is more and more widely used in the field of bio-informatics. By constructing a penalty function, it compresses certain regression coefficients or making them zero, thereby establishing a more accurate model, especially for independent variables High-dimensional situation. In the model established in this study, survival time and status were taken as dependent variables, and immune gene pairs were taken as independent variables. We utilized the "glmnet" R language package (version 2.0-16) to fit the logistic LASSO regression model [13, 14]. As a result, risk score of each pair was calculated and summarized. Meanwhile, we obtained these immune-gene-pairs which have the most significant impact on prognosis. In addition, in order to verify the predictive effect of the model, our team drew ROC curve and got the cutoff value. Based on it, high and low risk groups were dived and survival curve was built. Combing clinical information with risk file, our study performed univariate and multivariate independent prognostic analysis. At the same time, we drew ROC curve, forest map and boxplot curve for visual observation. R language and several packages were applied during it, including “survival”, “survivalROC”, “limma” and “ggpubr”.
Key genes
Through lasso regression, most related immune-gene-pairs were selected and they were believed to play important roles in GBM patients. Similarly, gene-pairs composed of these genes were believed to play an important role in GBM patients. These genes were so-called key genes in this study. In addition to explore the role of gene-pairs, individual or co-expression of key genes in GBM patients were worthy of studying. By the way, violin program was used to visually represent the individual role and “***” meant p-value < 0.01, since we compare certain genes’ expression level between patients with different prognosis.
cBioPortal
The cBioPortal for Cancer Genomics (http://cbioportal.org) is an open and user-friendly dataset, which could explore and analyze multidimensional cancer genomics data [15]. By selecting GBM database and setting gene names, it is easy to interactively probe genetic alterations across samples, genes, and pathways. Due to its intuitiveness and simplicity of operation, the database has inspired many new discoveries in cancer bioinformatics. Through co-expression analysis, it suggests interesting pathways that maybe worth further exploration.
Immune databases
xCell (http://xCell.ucsf.edu/) shows a novel method to present immune cells infiltration, which could explore 64 immune and stromal cell types [16]. Timer ((https://cistrome.shinyapps.io/timer/) comprehensively collects and analyze molecular characterization, investigating expression levels of six immune-interaction subtypes in 10,897 tumors from 32 cancer types [17]. QuanTIseq (http://icbi.at/quantiseq) is a fantastic way to quantify immune-infiltration, measuring the fractions of ten immune cell subtypes from a great number of RNA-sequencing data [18]. MCP-counter (Microenvironment Cell Populations-counter) focuses on exploring the infiltration of eight types of immune cells, thereby exploring the tumor microenvironment [19]. As a prospective cohort study, The European Prospective Investigation into Cancer and Nutrition (EPIC) aimed at investigating relationship between tumor and nutrition [20]. CIBERSORT (http://cibersort.stanford.edu/) could analyze RNA mixtures for exploring cellular biomarkers and therapeutic targets by sequencing samples from fresh, frozen and fixed tissues [21].
Although results and ways of these immune databases are not all the same, systematic integration of the immune infiltration in multiple immune cells can provide us with more convincing evidence. R packages “limma”, “scales”, “ggplot2” and “ggtext” were utilized to extract specific types of immune cell expression level and evaluate its correlation with risk score.