Integrate low dimensional embeddings

Source: R/generics.R, R/integration.R

Perform dataset integration using a pre-computed Anchorset of specified low dimensional representations.

IntegrateEmbeddings(anchorset, ...)

# S3 method for IntegrationAnchorSet
IntegrateEmbeddings(
  anchorset,
  new.reduction.name = "integrated_dr",
  reductions = NULL,
  dims.to.integrate = NULL,
  k.weight = 100,
  weight.reduction = NULL,
  sd.weight = 1,
  sample.tree = NULL,
  preserve.order = FALSE,
  verbose = TRUE,
  ...
)

# S3 method for TransferAnchorSet
IntegrateEmbeddings(
  anchorset,
  reference,
  query,
  query.assay = NULL,
  new.reduction.name = "integrated_dr",
  reductions = "pcaproject",
  dims.to.integrate = NULL,
  k.weight = 100,
  weight.reduction = NULL,
  reuse.weights.matrix = TRUE,
  sd.weight = 1,
  preserve.order = FALSE,
  verbose = TRUE,
  ...
)

Arguments

anchorset

An AnchorSet object

...

Reserved for internal use

new.reduction.name

Name for new integrated dimensional reduction.

reductions

Name of reductions to be integrated. For a TransferAnchorSet, this should be the name of a reduction present in the anchorset object (for example, "pcaproject"). For an IntegrationAnchorSet, this should be a DimReduc object containing all cells present in the anchorset object.

dims.to.integrate

Number of dimensions to return integrated values for

k.weight

Number of neighbors to consider when weighting anchors

weight.reduction

Dimension reduction to use when calculating anchor weights. This can be one of:

  • A string, specifying the name of a dimension reduction present in all objects to be integrated

  • A vector of strings, specifying the name of a dimension reduction to use for each object to be integrated

  • A vector of DimReduc objects, specifying the object to use for each object in the integration

  • NULL, in which case the full corrected space is used for computing anchor weights.

sd.weight

Controls the bandwidth of the Gaussian kernel for weighting

sample.tree

Specify the order of integration. Order of integration should be encoded in a matrix, where each row represents one of the pairwise integration steps. Negative numbers specify a dataset, positive numbers specify the integration results from a given row (the format of the merge matrix included in the hclust function output). For example: matrix(c(-2, 1, -3, -1), ncol = 2) gives:

            [,1]  [,2]
       [1,]   -2   -3
       [2,]    1   -1

Which would cause dataset 2 and 3 to be integrated first, then the resulting object integrated with dataset 1.

If NULL, the sample tree will be computed automatically.

preserve.order

Do not reorder objects based on size for each pairwise integration.

verbose

Print progress bars and output

reference

Reference object used in anchorset construction

query

Query object used in anchorset construction

query.assay

Name of the Assay to use from query

reuse.weights.matrix

Can be used in conjunction with the store.weights parameter in TransferData to reuse a precomputed weights matrix.

Value

When called on a TransferAnchorSet (from FindTransferAnchors), this will return the query object with the integrated embeddings stored in a new reduction. When called on an IntegrationAnchorSet (from IntegrateData), this will return a merged object with the integrated reduction stored.

Details

The main steps of this procedure are identical to IntegrateData with one key distinction. When computing the weights matrix, the distance calculations are performed in the full space of integrated embeddings when integrating more than two datasets, as opposed to a reduced PCA space which is the default behavior in IntegrateData.