如何使用ggbiplot处理pcaRes对象？绘制含缺失值数据的PCA结果

Got it! Since prcomp() can’t handle missing values directly, and ggbiplot is built to work with prcomp/princomp objects, we’ll use FactoMineR for PCA on data with missing values, then convert its pcaRes output into a format ggbiplot understands. Here’s a step-by-step guide tailored to your workflow:

1. Run PCA on Data with Missing Values using FactoMineR #

FactoMineR’s PCA() function natively supports missing value imputation (mean, median, or EM algorithm). Let’s use a modified iris dataset with simulated missing values as an example:

# Install/load required packages
install.packages("FactoMineR")
library(FactoMineR)
library(ggbiplot)

# Create iris dataset with missing values (simulate real-world scenario)
data(iris)
set.seed(123)
iris_missing <- iris
# Randomly insert 15 missing values across 2 numeric columns
iris_missing[sample(nrow(iris), 15), sample(1:4, 2)] <- NA

# Run PCA with missing value handling
pcaRes <- PCA(iris_missing[, 1:4],
              center = TRUE, scale.unit = TRUE,
              na.method = "mean", # Use mean imputation; try "EM" for better accuracy with more missing data
              graph = FALSE) # Skip FactoMineR's default plot

2. Convert pcaRes to a ggbiplot-Compatible Format #

ggbiplot expects the structure of a prcomp object, so we’ll extract the key components from pcaRes and wrap them into a list with the right class:

# Build a prcomp-like object
pca_compatible <- list(
  x = pcaRes$ind$coord,        # Sample principal component scores
  rotation = pcaRes$var$coord, # Variable loadings
  sdev = sqrt(pcaRes$eig[, 1]),# Standard deviations of principal components
  center = pcaRes$call$center, # Centering flag (matches your prcomp setup)
  scale = pcaRes$call$scale.unit # Scaling flag
)

# Assign the prcomp class so ggbiplot recognizes it
class(pca_compatible) <- "prcomp"

3. Plot with ggbiplot (Just Like You're Used To!) #

Now you can use your familiar ggbiplot syntax to create polished visuals—including grouping, ellipses, variable axes, and transparency:

# Generate the plot
p <- ggbiplot(pca_compatible,
              obs.scale = 1, var.scale = 1,
              ellipse = TRUE, circle = FALSE,
              varname.size = 3, var.axes = TRUE,
              groups = iris_missing$Species,
              alpha = 0.6) # Adjust transparency to highlight group overlap

# Add a clean theme and title
p <- p + theme_minimal() + ggtitle("PCA of Iris Data with Missing Values (Imputed)")

# Display the plot
print(p)

Quick Notes #

• For missing value handling: na.method="EM" uses the Expectation-Maximization algorithm, which is more robust than mean/median imputation if you have a lot of missing data.
• If your pcaRes comes from another package, the core idea stays the same: extract sample scores, variable loadings, and principal component standard deviations to build a prcomp-style object.

内容的提问来源于stack exchange，提问作者DaniCee