MetaNeighbor: Runs MetaNeighbor
In MetaNeighbor: Single cell replicability analysis

Description Usage Arguments Value See Also Examples

For each gene set of interest, the function builds a network of rank correlations between all cells. Next,It builds a network of rank correlations between all cells for a gene set. Next, the neighbor voting predictor produces a weighted matrix of predicted labels by performing matrix multiplication between the network and the binary vector indicating cell type membership, then dividing each element by the null predictor (i.e., node degree). That is, each cell is given a score equal to the fraction of its neighbors (including itself), which are part of a given cell type. For cross-validation, we permute through all possible combinations of leave-one-dataset-out cross-validation, and we report how well we can recover cells of the same type as area under the receiver operator characteristic curve (AUROC). This is repeated for all folds of cross-validation, and the mean AUROC across folds is reported. Calls neighborVoting.

MetaNeighbor(
  dat,
  i = 1,
  experiment_labels,
  celltype_labels,
  genesets,
  bplot = TRUE,
  fast_version = FALSE,
  node_degree_normalization = TRUE,
  batch_size = 10,
  detailed_results = FALSE
)

`dat`	A SummarizedExperiment object containing gene-by-sample expression matrix.
`i`	default value 1; non-zero index value of assay containing the matrix data
`experiment_labels`	A vector that indicates the source/dataset of each sample.
`celltype_labels`	A character vector or one-hot encoded matrix (cells x cell type) that indicates the cell type of each sample.
`genesets`	Gene sets of interest provided as a list of vectors.
`bplot`	default true, beanplot is generated
`fast_version`	default value FALSE; a boolean flag indicating whether to use the fast and low memory version of MetaNeighbor
`node_degree_normalization`	default value TRUE; a boolean flag indicating whether to normalize votes by dividing through total node degree.
`batch_size`	Optimization parameter. Gene sets are processed in groups of size batch_size. The count matrix is first subset to all genes from these groups, then to each gene set individually.
`detailed_results`	Should the function return the average AUROC across all test datasets (default) or a detailed table with the AUROC for each test dataset?

A matrix of AUROC scores representing the mean for each gene set tested for each celltype is returned directly (see neighborVoting). If detailed_results is set to TRUE, the function returns a table of AUROC scores in each test dataset for each gene set.

neighborVoting

data("mn_data")
data("GOmouse")
library(SummarizedExperiment)
AUROC_scores = MetaNeighbor(dat = mn_data,
                            experiment_labels = as.numeric(factor(mn_data$study_id)),
                            celltype_labels = metadata(colData(mn_data))[["cell_labels"]],
                            genesets = GOmouse,
                            bplot = TRUE)