rliger (version 1.0.0)

quantile_norm: Quantile align (normalize) factor loadings

Description

This process builds a shared factor neighborhood graph to jointly cluster cells, then quantile normalizes corresponding clusters.

Usage

quantile_norm(object, ...)
# S3 method for list
quantile_norm(
  object,
  quantiles = 50,
  ref_dataset = NULL,
  min_cells = 20,
  knn_k = 20,
  dims.use = NULL,
  do.center = FALSE,
  max_sample = 1000,
  eps = 0.9,
  refine.knn = TRUE,
  rand.seed = 1,
  ...
)
# S3 method for liger
quantile_norm(
  object,
  quantiles = 50,
  ref_dataset = NULL,
  min_cells = 20,
  knn_k = 20,
  dims.use = NULL,
  do.center = FALSE,
  max_sample = 1000,
  eps = 0.9,
  refine.knn = TRUE,
  rand.seed = 1,
  ...
)

Arguments

object

liger object. Should run optimizeALS before calling.

...

Arguments passed to other methods

quantiles

Number of quantiles to use for quantile normalization (default 50).

ref_dataset

Name of dataset to use as a "reference" for normalization. By default, the dataset with the largest number of cells is used.

min_cells

Minimum number of cells to consider a cluster shared across datasets (default 20)

knn_k

Number of nearest neighbors for within-dataset knn graph (default 20).

dims.use

Indices of factors to use for shared nearest factor determination (default 1:ncol(H[[1]])).

do.center

Centers the data when scaling factors (useful for less sparse modalities like methylation data). (default FALSE)

max_sample

Maximum number of cells used for quantile normalization of each cluster and factor. (default 1000)

eps

The error bound of the nearest neighbor search. (default 0.9) Lower values give more accurate nearest neighbor graphs but take much longer to computer.

refine.knn

whether to increase robustness of cluster assignments using KNN graph.(default TRUE)

rand.seed

Random seed to allow reproducible results (default 1)

Value

liger object with 'H.norm' and 'clusters' slot set.

Details

The first step, building the shared factor neighborhood graph, is performed in SNF(), and produces a graph representation where edge weights between cells (across all datasets) correspond to their similarity in the shared factor neighborhood space. An important parameter here is knn_k, the number of neighbors used to build the shared factor space.

Next we perform quantile alignment for each dataset, factor, and cluster (by stretching/compressing datasets' quantiles to better match those of the reference dataset). These aligned factor loadings are combined into a single matrix and returned as H.norm.

Examples

# NOT RUN {
# ligerex (liger object), factorization complete
# do basic quantile alignment
ligerex <- quantile_norm(ligerex)
# higher resolution for more clusters (note that SNF is conserved)
ligerex <- quantile_norm(ligerex, resolution = 1.2)
# change knn_k for more fine-grained local clustering
ligerex <- quantile_norm(ligerex, knn_k = 15, resolution = 1.2)
# }