Fit DeepInsight Feature Layout

Uses t-SNE on feature profiles (each feature viewed across samples) to project features to 2D positions. Discretizes to a grid. Resolves grid collisions by nearest-free-cell search.

Must be FIT on training data only to avoid leakage.

Requires Rtsne package.

Usage

fit_deepinsight(X_train, grid_size = 14L, perplexity = NULL, seed = 42L)

Arguments

X_train

Numeric matrix (n_train x p features)

grid_size

Integer, size of the square output grid (default: 14)

perplexity

t-SNE perplexity; if NULL, auto-chosen as min(5, p-1)

seed

Random seed for reproducibility

Value

A list with:

• positions: p x 2 integer matrix of (row, col) per feature

• grid_size: the grid dimension

References

Sharma A et al. (2019). Scientific Reports 9:11399.

Examples

if (FALSE) { # \dontrun{
if (requireNamespace("Rtsne", quietly = TRUE)) {
  X_train <- matrix(rnorm(70), nrow = 10, ncol = 7)
  layout <- fit_deepinsight(X_train, grid_size = 7, seed = 1)
}
} # }