Principal Component Analysis

Perform principal component analysis (PCA) on a proteomic data matrix, via the decomposition of the variance-covariance matrix (SVD). Some modification of the standard prcomp() is performed. Data transformations (e.g. centering and/or scaling) can be performed via its arguments.

The plot_scree() function generates a "scree plot" of a pca class object.

Usage

pca(data, features = NULL, center = TRUE, scale = FALSE)

# S3 method for class 'pca'
print(x, ...)

# S3 method for class 'pca'
plot(
  x,
  type = c("projection", "rotation"),
  dims = 1:2L,
  color,
  identify = FALSE,
  id.labels,
  ...
)

plot_scree(
  x,
  n = min(15L, length(attributes(x)$sdev)),
  type = c("barplot", "lines")
)

Arguments

data: A data.frame (or tbl_df) class object.
features: character(n). Which columns to consider the features.
center: logical(1). Should the features/variables be zero centered prior to decomposition? In general this should be performed so that variables are in the same space, thus defaults to TRUE.
scale: logical(1). Should the features be scaled to have unit variance prior to decomposition? If TRUE, this corresponds to PCA in correlation space as opposed to covariance space.
x: A pca class object.
...: Additional arguments passed to geom_point().
type: Either "projection" (default) or "rotation". If called from plot_scree(), either "barplot" (default) or "lines". Will be matched.
dims: integer(2). Which dimensions to plot.
color: An unquoted string indicating the variable in x$projection to color points by. Required for projection plots.
identify: Logical. Identify points? For rotation space, the top 10 target names in each dimension are identified.
id.labels: An unquoted string indicating the variable in x$projection containing point labels. Ignored for rotation plots.
n: integer(1). The number of components to plot.

Value

A pca class object, similar to prcomp().

Methods (by generic)

print(pca): The S3 print method for pca class objects.
plot(pca): The S3 plot method for pca class objects.

Author

Stu Field

Examples

pca <- pca(pcapkg:::log10_ft(simdata))

# print method
pca
#> ── PCA Object ─────────────────────────────────────────────────────────
#> Rotation                 40 x 49 (PC1:PC40)
#> Projection               100 x 56 (PC1:PC40)
#> Projection variables     '.id', 'id', 'class_response', 'reg_response', 'time', 'status', 'SampleId', 'SlideId', 'Subarray', 'SiteId', 'age', 'gender', 'HybControlNormScale', 'NormScale_40', 'NormScale_0_005', 'NormScale_1'
#> ═══════════════════════════════════════════════════════════════════════

# S3 plot methods
plot(pca, "r")                         # default rotation

plot(pca, "r", identify = TRUE)        # label targets


plot(pca, "p", color = class_response) # default samples projection

plot(pca, "p", color = gender)         # use "gender" for colors

plot(pca, "p", color = class_response, id.labels = id) # label points

plot(pca, "p", color = class_response, alpha = 0.5)    # adjust point alpha
#> Warning: Duplicated aesthetics after name standardisation: alpha

# Scree plots
plot_scree(pca)               # barplot

plot_scree(pca, type = "l")   # lines

plot_scree(pca, type = "l")