---
title: "GeoTcgaData"
output: rmarkdown::html_vignette
vignette: >
%\VignetteIndexEntry{GeoTcgaData}
%\VignetteEngine{knitr::rmarkdown}
%\VignetteEncoding{UTF-8}
---
```{r, include = FALSE}
knitr::opts_chunk$set(
collapse = TRUE,
comment = "#>"
)
```
--------
## Authors
Erqiang Hu
Department of Bioinformatics, School of Basic Medical Sciences,
Southern Medical University.
## Introduction
GEO and TCGA provide us with a wealth of data, such as RNA-seq, DNA Methylation,
single nucleotide Variation and Copy number variation data.
It's easy to download data from TCGA using the gdc tool or `TCGAbiolinks`,
and some software provides organized TCGA data, such as
[UCSC Xena](http://xena.ucsc.edu/) , UCSCXenaTools, and
[sangerbox](http://vip.sangerbox.com/), but processing these data into a format
suitable for bioinformatics analysis requires more work. This R package was
developed to handle these data.
```{r setup}
library(GeoTcgaData)
```
## Example
This is a basic example which shows you how to solve a common problem:
### RNA-seq data differential expression analysis
It is convenient to use TCGAbiolinks or
[`GDCRNATools`](https://bioconductor.org/packages/GDCRNATools/) to download
and analysis Gene expression data. `TCGAbiolinks` use `edgeR` package to do
differential expression analysis, while `GDCRNATools` can implement three most
commonly used methods: limma, edgeR , and DESeq2 to identify differentially
expressed genes (DEGs).
Alicia Oshlack et al. claimed that unlike the chip data,
the RNA-seq data had one [bias](https://pubmed.ncbi.nlm.nih.gov/20132535/)[1]:
the larger the transcript length / mean read count , the more likely it was to
be identified as a differential gene,
while there was no such trend in the
[chip data](https://pubmed.ncbi.nlm.nih.gov/19371405/)[2].
However, when we use their chip data for difference analysis
(using the limma package), we find that chip data has the same trend as
RNA-seq data. And we also found this trend in the difference analysis results
given by the data
[authors](https://genome.cshlp.org/content/18/9/1509.long)[3].
It is worse noting that only technical replicate data, which has small gene
dispersions, shows this [bias](https://pubmed.ncbi.nlm.nih.gov/28545404/)[4].
This is because in technical replicate RNA-seq data a long gene has more
reads mapping to it compared to a short gene of similar expression,
and most of the statistical methods used to detect differential expression
have stronger detection ability for genes with more reads. However, we have
not deduced why there is such a bias in the current difference
analysis algorithms.
Some software, such as [CQN](http://www.bioconductor.org/packages/cqn/) ,
present a
[normalization algorithm](https://pubmed.ncbi.nlm.nih.gov/22285995/) [5]
to correct systematic biases(gene length bias and
[GC-content bias](https://pubmed.ncbi.nlm.nih.gov/22177264/)[6].
But they did not provide sufficient evidence to prove that the correction is
effective. We use the
[Marioni dataset](https://pubmed.ncbi.nlm.nih.gov/19371405/)[2] to verify the
correction effect of CQN and find that there is still a deviation
after correction:
[GOseq](http://bioconductor.org/packages/goseq/) [1]based on
Wallenius' noncentral hypergeometric distribution can effectively correct the
gene length deviation in enrichment analysis. However, the current RNA-seq data
often have no gene length bias, but only the expression amount(read count)
bias, GOseq may overcorrect these data, correcting originally unbiased data
into reverse bias.
GOseq also fails to correct for expression bias, therefore, read count bias
correction is still a challenge for us.
```{r, message=FALSE, warning=FALSE}
# use user-defined data
df <- matrix(rnbinom(400, mu = 4, size = 10), 25, 16)
df <- as.data.frame(df)
rownames(df) <- paste0("gene", 1:25)
colnames(df) <- paste0("sample", 1:16)
group <- sample(c("group1", "group2"), 16, replace = TRUE)
result <- differential_RNA(counts = df, group = group,
filte = FALSE, method = "Wilcoxon")
# use SummarizedExperiment object input
df <- matrix(rnbinom(400, mu = 4, size = 10), 25, 16)
rownames(df) <- paste0("gene", 1:25)
colnames(df) <- paste0("sample", 1:16)
group <- sample(c("group1", "group2"), 16, replace = TRUE)
nrows <- 200; ncols <- 20
counts <- matrix(
runif(nrows * ncols, 1, 1e4), nrows,
dimnames = list(paste0("cg",1:200),paste0("S",1:20))
)
colData <- S4Vectors::DataFrame(
row.names = paste0("sample", 1:16),
group = group
)
data <- SummarizedExperiment::SummarizedExperiment(
assays=S4Vectors::SimpleList(counts=df),
colData = colData)
result <- differential_RNA(counts = data, groupCol = "group",
filte = FALSE, method = "Wilcoxon")
```
### DNA Methylation data integration
use `TCGAbiolinks` data.
The codes may need to be modified if `TCGAbiolinks` updates.
So please read its documents.
```{r, message=FALSE, warning=FALSE}
# use user defined data
library(ChAMP)
cpgData <- matrix(runif(2000), nrow = 200, ncol = 10)
rownames(cpgData) <- paste0("cpg", seq_len(200))
colnames(cpgData) <- paste0("sample", seq_len(10))
sampleGroup <- c(rep("group1", 5), rep("group2", 5))
names(sampleGroup) <- colnames(cpgData)
cpg2gene <- data.frame(cpg = rownames(cpgData),
gene = rep(paste0("gene", seq_len(20)), 10))
result <- differential_methy(cpgData, sampleGroup,
cpg2gene = cpg2gene, normMethod = NULL)
# use SummarizedExperiment object input
library(ChAMP)
cpgData <- matrix(runif(2000), nrow = 200, ncol = 10)
rownames(cpgData) <- paste0("cpg", seq_len(200))
colnames(cpgData) <- paste0("sample", seq_len(10))
sampleGroup <- c(rep("group1", 5), rep("group2", 5))
names(sampleGroup) <- colnames(cpgData)
cpg2gene <- data.frame(cpg = rownames(cpgData),
gene = rep(paste0("gene", seq_len(20)), 10))
colData <- S4Vectors::DataFrame(
row.names = colnames(cpgData),
group = sampleGroup
)
data <- SummarizedExperiment::SummarizedExperiment(
assays=S4Vectors::SimpleList(counts=cpgData),
colData = colData)
result <- differential_methy(cpgData = data,
groupCol = "group", normMethod = NULL,
cpg2gene = cpg2gene)
```
**Note:** `ChAMP`has a large number of dependent packages.
If you cannot install it successfully, you can download each dependent package
separately(Source or Binary) and install it locally.
We provide two models to get methylation difference genes:
if model = "cpg", step1: calculate difference cpgs;
step2: calculate difference genes;
if model = "gene", step1: calculate the methylation level of genes;
step2: calculate difference genes.
We find that only model = "gene" has no deviation of CpG number.
### Copy number variation data integration and differential gene extraction
use TCGAbiolinks to download TCGA data(Gene Level Copy Number Scores)
```{r, message=FALSE, warning=FALSE}
# use random data as example
aa <- matrix(sample(c(0, 1, -1), 200, replace = TRUE), 25, 8)
rownames(aa) <- paste0("gene", 1:25)
colnames(aa) <- paste0("sample", 1:8)
sampleGroup <- sample(c("A", "B"), ncol(aa), replace = TRUE)
diffCnv <- differential_CNV(aa, sampleGroup)
```
### Difference analysis of single nucleotide Variation data
We provide SNP_QC function to do quality control of SNP data
```{r, message=FALSE, warning=FALSE}
snpDf <- matrix(sample(c("AA", "Aa", "aa"), 100, replace = TRUE), 10, 10)
snpDf <- as.data.frame(snpDf)
sampleGroup <- sample(c("A", "B"), 10, replace = TRUE)
result <- SNP_QC(snpDf)
```
Then use differential_SNP to do differential analysis.
```{r, message=FALSE, warning=FALSE}
#' snpDf <- matrix(sample(c("mutation", NA), 100, replace = TRUE), 10, 10)
#' snpDf <- as.data.frame(snpDf)
#' sampleGroup <- sample(c("A", "B"), 10, replace = TRUE)
#' result <- differential_SNP(snpDf, sampleGroup)
```
### GEO chip data processing
The function `gene_ave` could average the expression data of different
ids for the same gene in the GEO chip data. For example:
```{r, message=FALSE, warning=FALSE}
aa <- c("MARCH1","MARC1","MARCH1","MARCH1","MARCH1")
bb <- c(2.969058399,4.722410064,8.165514853,8.24243893,8.60815086)
cc <- c(3.969058399,5.722410064,7.165514853,6.24243893,7.60815086)
file_gene_ave <- data.frame(aa=aa,bb=bb,cc=cc)
colnames(file_gene_ave) <- c("Gene", "GSM1629982", "GSM1629983")
result <- gene_ave(file_gene_ave, 1)
```
Multiple genes symbols may correspond to a same chip id. The result of
function `repAssign` is to assign the expression of this id to each gene,
and function `repRemove` deletes the expression. For example:
```{r}
aa <- c("MARCH1 /// MMA","MARC1","MARCH2 /// MARCH3",
"MARCH3 /// MARCH4","MARCH1")
bb <- c("2.969058399","4.722410064","8.165514853","8.24243893","8.60815086")
cc <- c("3.969058399","5.722410064","7.165514853","6.24243893","7.60815086")
input_file <- data.frame(aa=aa,bb=bb,cc=cc)
repAssign_result <- repAssign(input_file," /// ")
repRemove_result <- repRemove(input_file," /// ")
```
### Other downstream analyses
1. The function `id_conversion_TCGA` could convert ENSEMBL gene id to
gene Symbol in TCGA. For example:
```{r, message=FALSE, warning=FALSE}
data(profile)
result <- id_conversion_TCGA(profile)
```
The parameter `profile` is a data.frame or matrix of gene expression
data in TCGA.
**Note:** In previous versions(< 1.0.0) the `id_conversion` and
`id_conversion_TCGA` used HGNC data to convert human gene id.
In future versions, we will use `clusterProfiler::bitr` for ID conversion.
2. The function `countToFpkm` and `countToTpm` could convert
count data to FPKM or TPM data.
```{r}
data(gene_cov)
lung_squ_count2 <- matrix(c(1,2,3,4,5,6,7,8,9),ncol=3)
rownames(lung_squ_count2) <- c("DISC1","TCOF1","SPPL3")
colnames(lung_squ_count2) <- c("sample1","sample2","sample3")
result <- countToFpkm(lung_squ_count2, keyType = "SYMBOL",
gene_cov = gene_cov)
```
```{r, message=FALSE, warning=FALSE}
data(gene_cov)
lung_squ_count2 <- matrix(c(0.11,0.22,0.43,0.14,0.875,
0.66,0.77,0.18,0.29),ncol=3)
rownames(lung_squ_count2) <- c("DISC1","TCOF1","SPPL3")
colnames(lung_squ_count2) <- c("sample1","sample2","sample3")
result <- countToTpm(lung_squ_count2, keyType = "SYMBOL",
gene_cov = gene_cov)
```
```{r}
sessionInfo()
```
## References
1. Young MD, Wakefield MJ, Smyth GK, Oshlack A (2010) Gene ontology analysis
for RNA-seq: accounting for selection bias. Genome Biol 11: R14.
2. Oshlack A, Wakefield MJ (2009) Transcript length bias in RNA-seq data
confounds systems biology. Biol Direct 4: 14.
3. Marioni JC, Mason CE, Mane SM, Stephens M, Gilad Y (2008) RNA-seq: an
assessment of technical reproducibility and comparison with gene expression
arrays. Genome Res 18: 1509-1517.
4. Yoon S, Nam D (2017) Gene dispersion is the key determinant of the read
count bias in differential expression analysis of RNA-seq data.
BMC Genomics 18: 408.
5. Hansen KD, Irizarry RA, Wu Z (2012) Removing technical variability in
RNA-seq data using conditional quantile normalization.
Biostatistics 13: 204-216.
6. Risso D, Schwartz K, Sherlock G, Dudoit S (2011) GC-content normalization
for RNA-Seq data. BMC Bioinformatics 12: 480.