Skip to contents

[Experimental] Align factor loading by centroid alignment (beta)

Source: R/integration.R
centroidAlign.Rd

This process treats the factor loading of each dataset as the low dimensional embedding as well as the cluster assignment probability, i.e. the soft clustering result. Then the method aligns the embedding by linearly moving the centroids of the same cluster but within each dataset towards each other.

ATTENTION: This method is still under development while has shown encouraging results in benchmarking tests. The arguments and their default values reflect the best scored parameters in the tests and some of them may be subject to change in the future.

Usage

centroidAlign(object, ...)

# S3 method for liger
centroidAlign(
  object,
  lambda = 1,
  useDims = NULL,
  scaleEmb = TRUE,
  centerEmb = TRUE,
  scaleCluster = FALSE,
  centerCluster = FALSE,
  shift = FALSE,
  diagnosis = FALSE,
  ...
)

# S3 method for Seurat
centroidAlign(
  object,
  reduction = "inmf",
  lambda = 1,
  useDims = NULL,
  scaleEmb = TRUE,
  centerEmb = TRUE,
  scaleCluster = FALSE,
  centerCluster = FALSE,
  shift = FALSE,
  diagnosis = FALSE,
  ...
)

Arguments

object

A liger or Seurat object with valid factorization result available (i.e. runIntegration performed in advance).

...

Arguments passed to other S3 methods of this function.

lambda

Ridge regression penalty applied to each dataset. Can be one number that applies to all datasets, or a numeric vector with length equal to the number of datasets. Default 1.

useDims

Indices of factors to use considered for the alignment. Default NULL uses all factors.

scaleEmb

Logical, whether to scale the factor loading being considered as the embedding. Default TRUE.

centerEmb

Logical, whether to center the factor loading being considered as the embedding before scaling it. Default TRUE.

scaleCluster

Logical, whether to scale the factor loading being considered as the cluster assignment probability. Default FALSE.

centerCluster

Logical, whether to center the factor loading being considered as the cluster assignment probability before scaling it. Default FALSE.

shift

Logical, whether to shift the factor loading being considered as the cluster assignment probability after centered scaling. Default FALSE.

diagnosis

Logical, whether to return cell metadata variables with diagnostic information. See Details. Default FALSE.

reduction

Name of the reduction where LIGER integration result is stored. Default "inmf".

Value

Returns the updated input object

  • liger method

    • Update the H.norm slot for the aligned cell factor loading, ready for running graph based community detection clustering or dimensionality reduction for visualization.

    • Update the cellMata slot with diagnostic information if diagnosis = TRUE.

  • Seurat method

    • Update the reductions slot with a new DimReduc object containing the aligned cell factor loading.

    • Update the metadata with diagnostic information if diagnosis = TRUE.

Details

Diagnostic information include:

  • object$raw_which.max: The index of the factor with the maximum value in the raw factor loading.

  • object$R_which.max: The index of the factor with the maximum value in the soft clustering probability matrix used for correction.

  • object$Z_which.max: The index of the factor with the maximum value in the aligned factor loading.

Examples

pbmc <- centroidAlign(pbmcPlot)