Introduction

This vignette shows a complete workflow of the TCGAbiolinks package. The code is divided in 4 case study:

1. Expression pipeline (BRCA)
2. Expression pipeline (GBM)
3. Integration of DNA methylation and RNA expression pipeline (COAD)
4. ELMER pipeline (KIRC)

Case study n. 1: Pan Cancer downstream analysis BRCA

library(SummarizedExperiment)
library(TCGAbiolinks)
query.exp <- GDCquery(project = "TCGA-BRCA", 
legacy = TRUE,
data.category = "Gene expression",
data.type = "Gene expression quantification",
platform = "Illumina HiSeq", 
file.type = "results",
experimental.strategy = "RNA-Seq",
sample.type = c("Primary solid Tumor","Solid Tissue Normal"))
GDCdownload(query.exp)
brca.exp <- GDCprepare(query = query.exp, save = TRUE, save.filename = "brcaExp.rda")

# get subtype information
dataSubt <- TCGAquery_subtype(tumor = "BRCA")

# get clinical data
dataClin <- GDCquery_clinic(project = "TCGA-BRCA","clinical") 

# Which samples are primary solid tumor
dataSmTP <- TCGAquery_SampleTypes(getResults(query.exp,cols="cases"),"TP") 
# which samples are solid tissue normal
dataSmNT <- TCGAquery_SampleTypes(getResults(query.exp,cols="cases"),"NT")

Using TCGAnalyze_DEA, we identified 3,390 differentially expression genes (DEG) (log fold change >=1 and FDR < 1%) between 114 normal and 1097 BRCA samples. In order to understand the underlying biological process from DEGs we performed an enrichment analysis using TCGAnalyze_EA_complete function.

dataPrep <- TCGAanalyze_Preprocessing(object = brca.exp, cor.cut = 0.6)                      

dataNorm <- TCGAanalyze_Normalization(tabDF = dataPrep,
                                      geneInfo = geneInfo,
                                      method = "gcContent")                

dataFilt <- TCGAanalyze_Filtering(tabDF = dataNorm,
                                  method = "quantile", 
                                  qnt.cut =  0.25)   

dataDEGs <- TCGAanalyze_DEA(mat1 = dataFilt[,dataSmNT],
                            mat2 = dataFilt[,dataSmTP],
                            Cond1type = "Normal",
                            Cond2type = "Tumor",
                            fdr.cut = 0.01 ,
                            logFC.cut = 1,
                            method = "glmLRT")  

TCGAbiolinks outputs bar chart with the number of genes for the main categories of three ontologies (GO:biological process, GO:cellular component, and GO:molecular function, respectively).

ansEA <- TCGAanalyze_EAcomplete(TFname="DEA genes Normal Vs Tumor",
                                RegulonList = rownames(dataDEGs))  

TCGAvisualize_EAbarplot(tf = rownames(ansEA$ResBP),
                        GOBPTab = ansEA$ResBP,
                        GOCCTab = ansEA$ResCC,
                        GOMFTab = ansEA$ResMF,
                        PathTab = ansEA$ResPat,
                        nRGTab = rownames(dataDEGs),
                        nBar = 20)

The figure resulted from the code above is shown below.

The Kaplan-Meier analysis was used to compute survival univariate curves, and
log-Ratio test was computed to assess the statistical significance by using TCGAanalyze_SurvivalKM function; starting with 3,390 DEGs genes we found 555 significantly genes with p.value <0.05.

group1 <- TCGAquery_SampleTypes(colnames(dataFilt), typesample = c("NT"))
group2 <- TCGAquery_SampleTypes(colnames(dataFilt), typesample = c("TP"))

dataSurv <- TCGAanalyze_SurvivalKM(clinical_patient = dataClin,
                                   dataGE = dataFilt,
                                   Genelist = rownames(dataDEGs),
                                   Survresult = FALSE,
                                   ThreshTop = 0.67,
                                   ThreshDown = 0.33,
                                   p.cut = 0.05, group1, group2)

Cox-regression analysis was used to compute survival multivariate curves, and cox p-value was computed to assess the statistical significance by using TCGAnalyze_SurvivalCoxNET function. Survival multivariate analysis found 160 significantly genes according to the cox p-value FDR 5.00e-02. From DEGs that we found to correlate significantly with survival by both univariate and multivariate analyses we analyzed the following network.

The interactome network graph was generated using STRING.,org.Hs.string version 10 (Human functional protein association network). The network graph was resized by dnet package considering only multivariate survival genes, with strong interaction (threshold = 700) we obtained a subgraphsub graph of 24 nodes and 31 edges.

require(dnet)  # to change
org.Hs.string <- dRDataLoader(RData = "org.Hs.string")

TabCoxNet <- TCGAvisualize_SurvivalCoxNET(dataClin,
                                          dataFilt, 
                                          Genelist = rownames(dataSurv),
                                          scoreConfidence = 700,
                                          org.Hs.string = org.Hs.string,
                                          titlePlot = "Case Study n.1 dnet")

The figure resulted from the code above is shown below.

Case study n. 2: Pan Cancer downstream analysis LGG

We focused on the analysis of LGG samples. In particular, we used TCGAbiolinks to download 293 samples with molecular subtypes. Link the complete complete code. .

library(TCGAbiolinks)
library(SummarizedExperiment)

query.exp <- GDCquery(project = "TCGA-LGG", 
                      legacy = TRUE,
                      data.category = "Gene expression",
                      data.type = "Gene expression quantification",
                      platform = "Illumina HiSeq", 
                      file.type = "results",
                      experimental.strategy = "RNA-Seq",
                      sample.type = "Primary solid Tumor")
GDCdownload(query.exp)
lgg.exp <- GDCprepare(query = query.exp, save = TRUE, save.filename = "lggExp.rda")

First, we searched for possible outliers using the TCGAanalyze_Preprocessing function, which performs an Array Array Intensity correlation AAIC. We used all samples in expression data which contain molecular subtypes, filtering out samples without molecular information, and using only IDHmut-codel (n=85), IDHmut-non-codel (n=141) and IDHwt (n=56), NA (11), to define a square symetric matrix of pearson correlation among all samples (n=293). According to this matrix we found no samples with low correlation (cor.cut = 0.6) that can be identified as possible outliers, so we continued our analysis with 70 samples.

Second, using the TCGAanalyze_Normalization function we normalized mRNA transcripts and miRNA, using EDASeq package. This function does use Within-lane normalization procedures to adjust for GC-content effect (or other gene-level effects) on read counts: loess robust local regression, global-scaling, and full-quantile normalization (Risso et al. 2011) and between-lane normalization procedures to adjust for distributional differences between lanes (e.g., sequencing depth): global-scaling and full-quantile normalization (Bullard et al. 2010).

Third, using the TCGAanalyze_Filtering function we applied 3 filters removing features / mRNAs with low signal across samples obtaining 4578, 4284, 1187 mRNAs respectively.

Then we applied two Hierarchical cluster analysis on 1187 mRNAs after the three filters described above, the first cluster using as method ward.D2, and the second with ConsensusClusterPlus.

After the two clustering analysis, with cut.tree = 4 we obtained n= 4 espression clusters (EC).

library(dplyr)

dataPrep <- TCGAanalyze_Preprocessing(object = lgg.exp, cor.cut = 0.6)
dataNorm <- TCGAanalyze_Normalization(tabDF = dataPrep,
                                      geneInfo = geneInfo,
                                      method = "gcContent")

datFilt <- dataNorm %>% TCGAanalyze_Filtering(method = "varFilter") %>%
    TCGAanalyze_Filtering(method = "filter1") %>%  TCGAanalyze_Filtering(method = "filter2",foldChange = 0.2)

data_Hc2 <- TCGAanalyze_Clustering(tabDF = datFilt,
                                   method = "consensus",
                                   methodHC = "ward.D2") 
# Add  cluster information to Summarized Experiment
colData(lgg.exp)$groupsHC <- paste0("EC",data_Hc2[[4]]$consensusClass)

The next steps will be to visualize the data. First, we created the survival plot.

TCGAanalyze_survival(data = colData(lgg.exp),
                     clusterCol = "groupsHC",
                     main = "TCGA kaplan meier survival plot from consensus cluster",
                     legend = "RNA Group",height = 10,
                     risk.table = T,conf.int = F,
                     color = c("black","red","blue","green3"),
                     filename = "survival_lgg_expression_subtypes.png")

The result is showed below:

We will also, create a heatmap of the expression.

TCGAvisualize_Heatmap(t(datFilt),
                      col.metadata =  colData(lgg.exp)[,c("barcode",
                                                          "groupsHC",
                                                          "subtype_Histology",
                                                          "subtype_IDH.codel.subtype")],
                      col.colors =  list(
                          groupsHC = c("EC1"="black",
                                       "EC2"="red",
                                       "EC3"="blue",
                                       "EC4"="green3")),
                      sortCol = "groupsHC",
                      type = "expression", # sets default color
                      scale = "row", # use z-scores for better visualization. Center gene expression level around 0.
                      title = "Heatmap from concensus cluster", 
                      filename = "case2_Heatmap.png",
                      cluster_rows = TRUE,
                      color.levels = colorRampPalette(c("green", "black", "red"))(n = 11),
                      extrems =seq(-5,5,1),
                      cluster_columns = FALSE,
                      width = 1000,
                      height = 1000)

The result is shown below: