determine_ranks_tucker: Run rank determination by svd on the tensor unfolded along...

In scITD: Single-Cell Interpretable Tensor Decomposition

Run rank determination by svd on the tensor unfolded along each mode

Description

Run rank determination by svd on the tensor unfolded along each mode

Usage

determine_ranks_tucker(
  container,
  max_ranks_test,
  shuffle_level = "cells",
  shuffle_within = NULL,
  num_iter = 100,
  batch_var = NULL,
  norm_method = "trim",
  scale_factor = 10000,
  scale_var = TRUE,
  var_scale_power = 0.5,
  seed = container$experiment_params$rand_seed
)

Arguments

container	environment Project container that stores sub-containers for each cell type as well as results and plots from all analyses
max_ranks_test	numeric Vector of length 2 specifying the maximum number of donor and gene ranks to test
shuffle_level	character Either "cells" to shuffle cell-donor linkages or "tensor" to shuffle values within the tensor (default="cells")
shuffle_within	character A metadata variable to shuffle cell-donor linkages within (default=NULL)
num_iter	numeric Number of null iterations (default=100)
batch_var	character A batch variable from metadata to remove. No batch correction applied if NULL. (default=NULL)
norm_method	character The normalization method to use on the pseudobulked count data. Set to 'regular' to do standard normalization of dividing by library size. Set to 'trim' to use edgeR trim-mean normalization, whereby counts are divided by library size times a normalization factor. (default='trim')
scale_factor	numeric The number that gets multiplied by fractional counts during normalization of the pseudobulked data (default=10000)
scale_var	logical TRUE to scale the gene expression variance across donors for each cell type. If FALSE then all genes are scaled to unit variance across donors for each cell type. (default=TRUE)
var_scale_power	numeric Exponent of normalized variance that is used for variance scaling. Variance for each gene is initially set to unit variance across donors (for a given cell type). Variance for each gene is then scaled by multiplying the unit scaled values by each gene's normalized variance (where the effect of the mean-variance dependence is taken into account) to the exponent specified here. If NULL, uses var_scale_power from container$experiment_params. (default=.5)
seed	numeric Seed passed to set.seed() (default=container$experiment_params$rand_seed)

Value

The project container with a cowplot figure of rank determination plots in container$plots$rank_determination_plot.

Examples

test_container <- determine_ranks_tucker(test_container, max_ranks_test=c(3,5),
shuffle_level='tensor', num_iter=4, norm_method='trim', scale_factor=10000,
scale_var=TRUE, var_scale_power=.5)

scITD documentation built on Sept. 8, 2023, 5:11 p.m.