Perform UMAP Dimensionality Reduction

Run UMAP on the quantile normalized cell factors (result from quantileNorm), or unnormalized cell factors (result from runIntegration)) to generate a 2D embedding for visualization (or general dimensionality reduction). Has option to run on subset of factors. It is generally recommended to use this method for dimensionality reduction with extremely large datasets. The underlying UMAP calculation imports uwot umap.

Usage

runUMAP(
  object,
  useRaw = NULL,
  useDims = NULL,
  nDims = 2,
  distance = c("cosine", "euclidean", "manhattan", "hamming"),
  nNeighbors = 20,
  minDist = 0.1,
  dimredName = "UMAP",
  seed = 42,
  verbose = getOption("ligerVerbose", TRUE),
  k = nDims,
  use.raw = useRaw,
  dims.use = useDims,
  n_neighbors = nNeighbors,
  min_dist = minDist,
  rand.seed = seed
)

Arguments

object: liger object with factorization results.
useRaw: Whether to use un-aligned cell factor loadings (\(H\) matrices). Default NULL search for quantile-normalized loadings first and un-aligned loadings then.
useDims: Index of factors to use for computing UMAP embedding. Default NULL uses all factors.
nDims: Number of dimensions to reduce to. Default 2.
distance: Character. Metric used to measure distance in the input space. Default "cosine", alternative options include: "euclidean", "manhattan" and "hamming".
nNeighbors: Number of neighboring points used in local approximations of manifold structure. Default 10.
minDist: Numeric. Controls how tightly the embedding is allowed compress points together. Default 0.1.
dimredName: Name of the variable in cellMeta slot to store the result matrix. Default "UMAP".
seed: Random seed for reproducibility. Default 42.
verbose: Logical. Whether to show information of the progress. Default getOption("ligerVerbose") or TRUE if users have not set.
k, use.raw, dims.use, n_neighbors, min_dist, rand.seed: Deprecated. See Usage section for replacement.

Value

The object where a "UMAP" variable is updated in the cellMeta slot with the whole 2D embedding matrix.

Details

For nNeighbors, larger values will result in more global structure being preserved at the loss of detailed local structure. In general this parameter should often be in the range 5 to 50, with a choice of 10 to 15 being a sensible default.

For minDist, larger values ensure embedded points are more evenly distributed, while smaller values allow the algorithm to optimize more accurately with regard to local structure. Sensible values are in the range 0.001 to 0.5, with 0.1 being a reasonable default.

Examples