Calculate UMAP Dimensionality Reduction

Performs Uniform Manifold Approximation and Projection (UMAP) on the samples to generate a 2D embedding. This is a powerful non-linear method for visualizing sample relationships. This function serves as a wrapper for the umap::umap function.

Usage

get_umap(object, condition = NA, neighbors = 15, ...)

# S4 method for class 'SummarizedExperiment'
get_umap(object, condition, neighbors)

Arguments

object: A SummarizedExperiment object. The data should be imputed.
condition: A character string specifying the column name in the condition slot to use for labeling points. If NULL (the default), it will attempt to guess groups from sample names.
neighbors: The size of the local neighborhood UMAP will look at. This is a key hyperparameter affecting the balance between local and global structure. Defaults to 15.
...: Additional arguments passed on to the umap::umap function (e.g., min_dist, n_epochs, metric).

Value

A data.table with columns for Sample, UMAP1, UMAP2, and Condition.

Details

This function expects clean, imputed data. Missing values (NA) will cause an error. For meaningful results, it is highly recommended to use data that has been log-transformed and normalized.

Important: UMAP is a stochastic algorithm, meaning it will produce slightly different results each time it is run. For reproducible results, you must set a seed (e.g., set.seed(123)) before calling this function.

This function requires the umap package to be installed from CRAN.

Functions

get_umap(SummarizedExperiment): Method for SummarizedExperiment objects.