system_genetics/rnaseq/step6_overall_QC/01 - Principle Component Analysis.R

# Principle Component Analysis
# Normalized with limma::voom
library(limma)
library(tidyverse)


# expression.data. Has columns:
# - Gene (gene identifier)
# - [Sample identifiers]
expression.data <- "mrna_counts_table.csv"
# results.dir. The directory of where to put the resulting tables (and later your plots.)
results.dir <- "results"


# PCA variables
do.center = TRUE
do.scale = FALSE


# The analysis
# We use prcomp to calculate the PCAs. Afterwards you should plot the results.
norm.expr.data <- expression.data %>%
	tibble::column_to_rownames("Gene")
norm.expr.data <- norm.expr.data[rowSums(norm.expr.data) >= 10,] %>%
	limma::voom() %>%
	as.matrix()

# Principle Component analysis
results.dir.pca <- file.path(results.dir, "principle.components")
dir.create(results.dir.pca, recursive=TRUE)

norm.expr.data.pcs <- norm.expr.data %>%
  t() %>%
  stats::prcomp(
  	center = do.center,
  	scale. = do.scale
  )

# Write summary of PCAs to files
pcs.summary <- summary(norm.expr.data.pcs)
pcs.summary$importance %>%
	t() %>%
	as.data.frame() %>%
	tibble::rownames_to_column("PC.name") %>%
	readr::write_csv(
		file.path(results.dir.pca, "importance.csv")
	)

pcs.summary$x %>%
	t() %>%
	as.data.frame() %>%
	tibble::rownames_to_column("ensembl.id") %>%
	readr::write_csv(
		file.path(results.dir.pca, "values.csv")
	)

pcs.summary$rotation %>%
	t() %>%
	as.data.frame() %>%
	tibble::rownames_to_column("sample.id") %>%
	readr::write_csv(
		file.path(results.dir.pca, "rotation.csv")
	)

data.frame(
		rownames = names(pcs.summary$center),
		center = pcs.summary$center,
		scale = pcs.summary$scale
	) %>%
	readr::write_csv(
		file.path(results.dir.pca, "rest.csv")
	)

# Not saved: pcs.summary$sdev,


# Next thing to do:
# - (Optional) scree plot - to determine the optimal cutoff for PCA inclusion based on explaination of variance
# - (Optional) eigencorplot - to correlate PCAs to clinical variables so that you know which PCA to include for which analysis
# - (Optional) pairsplot - plot multiple PCAs against each other in a single figure
# - Plot the first couple of PCAs against each other