Graph-based feature discovery for landmarks

Identifies the most characteristic features (genes or markers) for each landmark by determining which features contribute most strongly to the landmark's defining principal component. This creates landmark-specific feature signatures useful for interpretation.

Usage

get.features(x, ...)

# S3 method for class 'TDRObj'
get.features(x, ...)

Arguments

x

A TDRObj, Seurat, SingleCellExperiment, or HDF5AnnData (anndataR) object.

...

Additional arguments passed to methods.

Value

Updated .tdr.obj with $interact.plot$lm.features containing:

• res: List of data frames (one per landmark) with top 10 feature importances

• html: HTML-formatted tables for interactive plotting

Feature importance is the signed loading from the landmark's top principal component.

Details

Algorithm:

1. For each landmark, identify the PC with highest absolute value (its "defining" PC)

2. Extract that PC's rotation vector (feature loadings)

3. Multiply by sign of landmark's PC score to preserve directionality

4. Rank features by signed loading; keep top 10

Positive loadings indicate features that increase along the PC direction, negative indicate features that decrease. The signed values reflect how the landmark is characterized relative to the population average.

Results are stored for use in plotPCA and plotUMAP, which displays feature signatures when hovering over landmarks when .hover.stats = "marker".

See also

plotPCA, plotUMAP, get.landmarks

Examples

if (FALSE) { # \dontrun{
# After landmark selection and graph construction
lm.cells <- setup.tdr.obj(.cells = .cells, .meta = .meta) |>
  get.landmarks() |>
  get.graph() |>
  get.features()

# View feature signature for first landmark
head(lm.cells$interact.plot$lm.features$res[[1]])

# Use in interactive plotting
plotPCA(lm.cells, .hover.stats = "marker")
} # }