Run UMAP

Runs the Uniform Manifold Approximation and Projection (UMAP) dimensional reduction technique. To run using umap.method="umap-learn", you must first install the umap-learn python package (e.g. via pip install umap-learn). Details on this package can be found here: https://github.com/lmcinnes/umap. For a more in depth discussion of the mathematics underlying UMAP, see the ArXiv paper here: https://arxiv.org/abs/1802.03426.

RunUMAP(object, ...)

# S3 method for default
RunUMAP(
  object,
  reduction.key = "UMAP_",
  assay = NULL,
  reduction.model = NULL,
  return.model = FALSE,
  umap.method = "uwot",
  n.neighbors = 30L,
  n.components = 2L,
  metric = "cosine",
  n.epochs = NULL,
  learning.rate = 1,
  min.dist = 0.3,
  spread = 1,
  set.op.mix.ratio = 1,
  local.connectivity = 1L,
  repulsion.strength = 1,
  negative.sample.rate = 5,
  a = NULL,
  b = NULL,
  uwot.sgd = FALSE,
  seed.use = 42,
  metric.kwds = NULL,
  angular.rp.forest = FALSE,
  densmap = FALSE,
  dens.lambda = 2,
  dens.frac = 0.3,
  dens.var.shift = 0.1,
  verbose = TRUE,
  ...
)

# S3 method for Graph
RunUMAP(
  object,
  assay = NULL,
  umap.method = "umap-learn",
  n.components = 2L,
  metric = "correlation",
  n.epochs = 0L,
  learning.rate = 1,
  min.dist = 0.3,
  spread = 1,
  repulsion.strength = 1,
  negative.sample.rate = 5L,
  a = NULL,
  b = NULL,
  uwot.sgd = FALSE,
  seed.use = 42L,
  metric.kwds = NULL,
  densmap = FALSE,
  densmap.kwds = NULL,
  verbose = TRUE,
  reduction.key = "UMAP_",
  ...
)

# S3 method for Neighbor
RunUMAP(object, reduction.model, ...)

# S3 method for Seurat
RunUMAP(
  object,
  dims = NULL,
  reduction = "pca",
  features = NULL,
  graph = NULL,
  assay = DefaultAssay(object = object),
  nn.name = NULL,
  slot = "data",
  umap.method = "uwot",
  reduction.model = NULL,
  return.model = FALSE,
  n.neighbors = 30L,
  n.components = 2L,
  metric = "cosine",
  n.epochs = NULL,
  learning.rate = 1,
  min.dist = 0.3,
  spread = 1,
  set.op.mix.ratio = 1,
  local.connectivity = 1L,
  repulsion.strength = 1,
  negative.sample.rate = 5L,
  a = NULL,
  b = NULL,
  uwot.sgd = FALSE,
  seed.use = 42L,
  metric.kwds = NULL,
  angular.rp.forest = FALSE,
  densmap = FALSE,
  dens.lambda = 2,
  dens.frac = 0.3,
  dens.var.shift = 0.1,
  verbose = TRUE,
  reduction.name = "umap",
  reduction.key = NULL,
  ...
)

Arguments

object

An object

...

Arguments passed to other methods and UMAP

reduction.key

dimensional reduction key, specifies the string before the number for the dimension names. UMAP by default

assay

Assay to pull data for when using features, or assay used to construct Graph if running UMAP on a Graph

reduction.model

DimReduc object that contains the umap model

return.model

whether UMAP will return the uwot model

umap.method

UMAP implementation to run. Can be

uwot:

Runs umap via the uwot R package

uwot-learn:

Runs umap via the uwot R package and return the learned umap model

umap-learn:

Run the Seurat wrapper of the python umap-learn package

n.neighbors

This determines the number of neighboring points used in local approximations of manifold structure. 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.

n.components

The dimension of the space to embed into.

metric

metric: This determines the choice of metric used to measure distance in the input space. A wide variety of metrics are already coded, and a user defined function can be passed as long as it has been JITd by numba.

n.epochs

he number of training epochs to be used in optimizing the low dimensional embedding. Larger values result in more accurate embeddings. If NULL is specified, a value will be selected based on the size of the input dataset (200 for large datasets, 500 for small).

learning.rate

The initial learning rate for the embedding optimization.

min.dist

This controls how tightly the embedding is allowed compress points together. Larger values ensure embedded points are moreevenly distributed, while smaller values allow the algorithm to optimise more accurately with regard to local structure. Sensible values are in the range 0.001 to 0.5.

spread

The effective scale of embedded points. In combination with min.dist this determines how clustered/clumped the embedded points are.

set.op.mix.ratio

Interpolate between (fuzzy) union and intersection as the set operation used to combine local fuzzy simplicial sets to obtain a global fuzzy simplicial sets. Both fuzzy set operations use the product t-norm. The value of this parameter should be between 0.0 and 1.0; a value of 1.0 will use a pure fuzzy union, while 0.0 will use a pure fuzzy intersection.

local.connectivity

The local connectivity required - i.e. the number of nearest neighbors that should be assumed to be connected at a local level. The higher this value the more connected the manifold becomes locally. In practice this should be not more than the local intrinsic dimension of the manifold.

repulsion.strength

Weighting applied to negative samples in low dimensional embedding optimization. Values higher than one will result in greater weight being given to negative samples.

negative.sample.rate

The number of negative samples to select per positive sample in the optimization process. Increasing this value will result in greater repulsive force being applied, greater optimization cost, but slightly more accuracy.

a

More specific parameters controlling the embedding. If NULL, these values are set automatically as determined by min. dist and spread. Parameter of differentiable approximation of right adjoint functor.

b

More specific parameters controlling the embedding. If NULL, these values are set automatically as determined by min. dist and spread. Parameter of differentiable approximation of right adjoint functor.

uwot.sgd

Set uwot::umap(fast_sgd = TRUE); see umap for more details

seed.use

Set a random seed. By default, sets the seed to 42. Setting NULL will not set a seed

metric.kwds

A dictionary of arguments to pass on to the metric, such as the p value for Minkowski distance. If NULL then no arguments are passed on.

angular.rp.forest

Whether to use an angular random projection forest to initialise the approximate nearest neighbor search. This can be faster, but is mostly on useful for metric that use an angular style distance such as cosine, correlation etc. In the case of those metrics angular forests will be chosen automatically.

densmap

Whether to use the density-augmented objective of densMAP. Turning on this option generates an embedding where the local densities are encouraged to be correlated with those in the original space. Parameters below with the prefix ‘dens’ further control the behavior of this extension. Default is FALSE. Only compatible with 'umap-learn' method and version of umap-learn >= 0.5.0

dens.lambda

Specific parameter which controls the regularization weight of the density correlation term in densMAP. Higher values prioritize density preservation over the UMAP objective, and vice versa for values closer to zero. Setting this parameter to zero is equivalent to running the original UMAP algorithm. Default value is 2.

dens.frac

Specific parameter which controls the fraction of epochs (between 0 and 1) where the density-augmented objective is used in densMAP. The first (1 - dens_frac) fraction of epochs optimize the original UMAP objective before introducing the density correlation term. Default is 0.3.

dens.var.shift

Specific parameter which specifies a small constant added to the variance of local radii in the embedding when calculating the density correlation objective to prevent numerical instability from dividing by a small number. Default is 0.1.

verbose

Controls verbosity

densmap.kwds

A dictionary of arguments to pass on to the densMAP optimization.

dims

Which dimensions to use as input features, used only if features is NULL

reduction

Which dimensional reduction (PCA or ICA) to use for the UMAP input. Default is PCA

features

If set, run UMAP on this subset of features (instead of running on a set of reduced dimensions). Not set (NULL) by default; dims must be NULL to run on features

graph

Name of graph on which to run UMAP

nn.name

Name of knn output on which to run UMAP

slot

The slot used to pull data for when using features. data slot is by default.

reduction.name

Name to store dimensional reduction under in the Seurat object

Value

Returns a Seurat object containing a UMAP representation

References

McInnes, L, Healy, J, UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction, ArXiv e-prints 1802.03426, 2018

Examples

if (FALSE) {
data("pbmc_small")
pbmc_small
# Run UMAP map on first 5 PCs
pbmc_small <- RunUMAP(object = pbmc_small, dims = 1:5)
# Plot results
DimPlot(object = pbmc_small, reduction = 'umap')
}