Practicing PCA

The goal of this lab is to have you all recreate the coding and analysis behind a typical Principal Component Analysis.

Please follow the tutorial linked here: https://bioinfo4all.wordpress.com/2021/01/31/tutorial-6-how-to-do-principal-component-analysis-pca-in-r/

This tutorial does a good job of describing why we use PCA and applies it to common data types for bioinformatics. As you go through the tutorial, copy and paste code below to perform the tasks. Additionally, I have a few questions you need to answer.

NOTE: I have already altered the .csv file you need for this, it is found on Canvas as “cell.csv”, bring in this file using the code below (your working directory will need to be set up correctly):

data <- read.csv("cell.csv", row.names = 1)  

The code above calls the dataframe “data”, if for some reason, you end up calling your data someting else, you will need to alter all of the rest of the code accordingly. Add code chunks below this starting the tutorial with the “install.packages” code on the website:

Load packages

library("factoextra")

## Loading required package: ggplot2

## Welcome to factoextra!

## Want to learn more? See two factoextra-related books at https://www.datanovia.com/library/principal-component-methods

library("FactoMineR")

PCA

pca.data <- PCA(data[,-1], scale.unit = TRUE, graph = FALSE)

#  Make sure most of the data will be present and plot to show with of the dimensions contributes to how much variation. 
fviz_eig(pca.data, addlabels = TRUE, ylim = c(0, 70))

## Warning: This FactoMineR PCA result contains only 5 eigenvalues and does not
## include the complete spectrum. Refit the PCA with a larger `ncp` before drawing
## a complete scree plot.

# Correlation plot
fviz_pca_var(pca.data, col.var = "cos2",
             gradient.cols = c("#FFCC00", "#CC9933", "#660033", "#330033"),
             repel = TRUE) 

Plot cell types

# Plot cell types and flip to put cell types as rows
pca.data <- PCA(t(data[,-1]), scale.unit = TRUE, graph = FALSE)

# Visualize
fviz_pca_ind(pca.data, col.ind = "cos2", 
                  gradient.cols = c("#FFCC00", "#CC9933", "#660033", "#330033"), 
                  repel = TRUE)

# Add labels
library(ggpubr) 
a <- fviz_pca_ind(pca.data, col.ind = "cos2", 
                  gradient.cols = c("#FFCC00", "#CC9933", "#660033", "#330033"), 
                  repel = TRUE)
ggpar(a,
      title = "Principal Component Analysis",
      xlab = "PC1", ylab = "PC2",
      legend.title = "Cos2", legend.position = "top",
      ggtheme = theme_minimal())

Plot gene expression

pca.data <- PCA(data[,-1], scale.unit = TRUE,ncp = 2, graph = FALSE)

# Convert lineage to a factor
data$lineage <- as.factor(data$lineage)

# Color palette 
library(RColorBrewer)
nb.cols <- 3
mycolors <- colorRampPalette(brewer.pal(3, "Set1"))(nb.cols) 

# Create PCA plot and add labels
 a <- fviz_pca_ind(pca.data, col.ind = data$lineage,
                  palette = mycolors, addEllipses = TRUE)
 
 ggpar(a,
      title = "Principal Component Analysis",
      xlab = "PC1", ylab = "PC2",
      legend.title = "Cell type", legend.position = "top",
      ggtheme = theme_minimal())

QUESTIONS: 1. Why does the author of this tutorial suggest PCA is a powerful analysis tool?

• PCA helps make sense of big data, gives the data shape, and identifies which samples are similar and different.

2. Why do you think this data makes sense for working with PCAs?

• This data is of gene expression measurements for many genes across different cells giving it many elements that would make it difficult to identify overall patterns without PCA.

3. How do you interpret the Scree plot?

• The scree plot tells us how much of the total variation can be explained be each principal component.

4. How do you interpret the final PCA plot you make?

• The final PCA visualizes the similarities and differences between samples based on gene expression profiles. Points close together are more similar and have similar genetic profiles, while points further apart are more different, with different genetic profiles.