Page Comparison

# Differential expression analysis

# When you see '## USER INPUT', this means you have to modify the code for your computer or dataset. All other code can be run as-is (i.e. you don't need to understand the code, just run it)

#### Installing required packages ####

# **NOTE: this section only needs to be run once (or occasionally to update the packages)
# Install devtools
install.packages("devtools", repos = "http://cran.us.r-project.org")
# Install R packages. This only needs to be run once.
bioconductor_packages <- c("DESeq2", "EnhancedVolcano")
cran_packages <- c("ggrepel", "ggplot2", "plyr", "reshape2", "readxl", "FactoMineR", "factoextra")
# Compares installed packages to above packages and returns a vector of missing packages
new_packages <- bioconductor_packages[!(bioconductor_packages %in% installed.packages()[,"Package"])]
new_cran_packages <- cran_packages[!(cran_packages %in% installed.packages()[,"Package"])]
# Install missing bioconductor packages
if (!requireNamespace("BiocManager", quietly = TRUE))
  install.packages("BiocManager")
BiocManager::install(new_packages)
# Install missing cran packages
if (length(new_cran_packages)) install.packages(new_cran_packages, repos = "http://cran.us.r-project.org")
# Update all installed packages are the latest version
update.packages(bioconductor_packages, ask = FALSE)
update.packages(cran_packages, ask = FALSE, repos = "http://cran.us.r-project.org")


#### Loading required packages ####

# This section needs to be run very time
# Load packages
bioconductor_packages <- c("DESeq2", "EnhancedVolcano")
cran_packages <- c("ggrepel", "ggplot2", "plyr", "reshape2", "readxl", "FactoMineR", "factoextra")
lapply(cran_packages, require, character.only = TRUE)
lapply(bioconductor_packages, require, character.only = TRUE)


#### Import your count data ####

# Make sure you have: a) your count table (salmon.merged.gene_counts.tsv file, if you used Nextflow nfcore/rnaseq to analyse your data). Copy this to a subdirectory called 'data'. b) your metadata file. This should be either an Excel file called 'metadata.xlsx' or a tab-separated text file called 'metadata.txt'. It needs 3 columns called 'sample_name', 'sample_ID' and 'group'. The sample names should be EXACTLY the same as the names in the count table. These names are often uninformative and long, so the 'sample_ID' is the sample labels you want to put on your plots. E.g. if you have a 'high fat' group, you might want to rename the samples HF1, HF2, HF3, etc)

## USER INPUT
# Set working directory. 
# Change this to your working directory (In the RStudio menu: Session -> Set working directory -> Choose working directory)
setwd("C:/Users/whatmorp/OneDrive - Queensland University of Technology/Desktop/Projects/RNA-Seq downstream analysis")

# Import your count data. make sure you've created a 'data' subdirectory and put the count table file there.
metacountdata <- read.table("./data/salmon.merged.gene_counts.tsv", header = TRUE, row.names = 1)

# Import metadata (option 1: text file). Again, need a metadata.txt file in the data subdirectory.
meta <- read.table("./data/metadata.txt", header = T)
# Import metadata (option 1: Excel file))
#meta <- read_excel("./data/metadata.xlsx")

# Remove 1st columns of metadata (gene_name)
counts <- metacountdata[ ,2:ncol(metacountdata)]

# Rename sample names to new sample IDs
counts <- counts[as.character(meta$sample_name)]
colnames(counts) <- meta$sample_ID

# Counts need to be rounded to integers
counts <- ceiling(counts)


#### Outliers and batch effects ####

# This section normalises and transforms the count data so that it can be plotted on a PCA plot and a heatmap

# Normalise counts by library size ##
# Using DeSeq2's estimateSizeFactors(). Note that DeSeq2 does this internally during DEG calling. The normalisation below is done separately for PCA and density plotting.

## USER INPUT
# Choose the groups you want to plot in a PCA/Heatmap. You can select any 2 or more of the groups (or all of the groups) you have in your 'groups' column of your metadata table
# To see what groups are present, run the following:
unique(meta$group)
# Now add which groups you want to plot (i.e. replace the groupnames below, and add more, separated by a comma and in "quotes", as needed). NOTE: R is case-sensitive, so these group names must be named EXACTLY the same as in the metadata table.
plotgroups <- c("Differentiated_cells", "Basal_cells")

# Pull out only the counts from the above groups
groupcounts <- counts[meta$group %in% plotgroups]

# Set up the initial DeSeq2 experimental parameters. This is the length of the no' of samples being analysed. In full DeSeq2 analysis this is to separate groups as factors, but here a simple count is fine.
condition <- factor(1:length(groupcounts))
# Set up the column data. A data frame of sample ID's and conditions
coldata <- data.frame(row.names=colnames(groupcounts), condition)
# Set up the DeSeq2 data set structure
f <- DESeqDataSetFromMatrix(countData = groupcounts, colData = coldata, design= ~ condition)
# Estimate the size factors. See DeSeq2 manual for details
f <- estimateSizeFactors(f)
# Size factors can be viewed by: sizeFactors(f)

# Multiply each row (sample) by the corresponding size factor
subcount_norm <- as.matrix(groupcounts) %*% diag(sizeFactors(f))

# Re-add column names
colnames(subcount_norm) <- colnames(groupcounts)

## Remove low coverage transcripts (mean count < 10) ##

# Find the mean of each row (and output as a data frame object)
means <- as.data.frame(rowMeans(subcount_norm))

# Then join the means data with the counts
means <- cbind(means, subcount_norm)

# Then subset out only genes with mean > 10
data <- subset(means, means[ , 1] > 10)

# Remove the means column
data <- data[,-1]

# Transform data
data_log <- vst(round(as.matrix(data)))

# Transformation can create some infinite values. Can't generate PCA data on these. Can see how many by: sum(sapply(data_log, is.infinite))
# To remove infinite rows, use 'is.finte' or '!is.infinite'
data_log <- data_log[is.finite(rowSums(data_log)),]
colnames(data_log) <- colnames(groupcounts)

### Set up the PCA plot base data ###

# We're using the FactoMineR package to generate PCA plots (http://factominer.free.fr/index.html)

# Need to transpose the data first
data_log_t <- t(data_log)

# Add the group data
data_log_t_vars <- data.frame(meta$group[meta$group %in% plotgroups], data_log_t)

# Generate the PCA data using FactoMineR package
res.pca <- PCA(data_log_t_vars, quali.sup = 1, graph=FALSE)

## Set up the dendogram/heatmaps base data ##

# Calculate the distance matrix:
distance_matrix <- as.matrix(dist(t(data_log)))


#### PCA plot ####

# Generate the PCA plot. Groups are shaded with elipses at 95% confidence level. NOTE: at least 4 replicates need to be in a group for an elipses to be drawn.
p <- fviz_pca_ind(res.pca,
                  geom.ind = "point", # show points only (but not "text")
                  col.ind = meta$group[meta$group %in% plotgroups], # color by groups
                  pointsize = 5, title = "",
                  addEllipses = TRUE, ellipse.type = "t", ellipse.level = 0.95) + theme(legend.text = element_text(size = 12), legend.title = element_text(size = 14), axis.title=element_text(size=16), axis.text=element_text(size=14))

p

## USER INPUT. Change these to 
pdf_filename <- "pca.pdf"
tiff_filename <- "pca.tiff"

# Output as publication quality (300dpi) tiff and pdf (Note: this will create a 'figures subdirectory where all figures will be output)
dir.create("figures", showWarnings = FALSE)
ggsave(file = paste0("./figures/", tiff_filename), dpi = 300, compression = "lzw", device = "tiff", width = 10, height = 8, plot = p)
ggsave(file = paste0("./figures/", pdf_filename), device = "pdf", width = 10, height = 8, plot = p)


#### Heatmap and dendrogram ####