4.9 Principal Component Analysis (PCA)

PCA reduces the high-dimensional expression space (one dimension per gene) to a small number of principal components that capture the largest sources of variance in the data. In a well-controlled experiment, PC1 should separate the two conditions — this confirms that the biological effect of interest is the dominant driver of transcriptional variation.

If PC1 is instead explained by a technical variable (batch, sequencing run, RNA quality), batch correction will be needed before differential expression analysis (Leek et al., 2010).

pca_data <- plotPCA(vsd,
                    intgroup  = c("sample", "condition"),
                    returnData = TRUE)

pct_var <- round(100 * attr(pca_data, "percentVar"), 1)

PCAPlot <- ggplot(pca_data, aes(x     = PC1,
                                y     = PC2,
                                color = condition,
                                label = sample)) +
  geom_point(size = 4) +
  geom_text(vjust = -0.8, size = 3, show.legend = FALSE) +
  geom_hline(yintercept = 0, linetype = "dashed", alpha = 0.3) +
  geom_vline(xintercept = 0, linetype = "dashed", alpha = 0.3) +
  theme_pubr(border = TRUE) +
  theme(
    axis.text       = element_text(size = 12),
    axis.title      = element_text(size = 14),
    legend.text     = element_text(size = 12),
    legend.position = "bottom"
  ) +
  labs(
    x     = paste0("PC1: ", pct_var[1], "% variance"),
    y     = paste0("PC2: ", pct_var[2], "% variance"),
    title = "PCA — E. coli MG1655 (VST-transformed counts)",
    color = "Condition"
  ) +
  scale_color_manual(values = cols_condition)

PCAPlot

ggsave(
  filename = file.path(git_root, "results", "plots", "PCA_Plot.png"),
  plot     = PCAPlot,
  width    = 8,
  height   = 6,
  dpi      = 300
)

💡 Extra: You can always use plotly!

ggplotly(PCAPlot)

PC1 capturing 85.4% and separating conditions cleanly is a strong positive signal — the saccharin treatment has a large, consistent transcriptional effect. C3 separating on PC2 is worth noting: it passed normalization QC but shows some residual transcriptional difference from C1/C2. This is within acceptable range for biological replicates, but we need to document it and check if C3 was grown or processed differently.