buildReference {symphony}

Function for building a Symphony reference starting from expression matrix

Description

Function for building a Symphony reference starting from expression matrix

Usage

buildReference(
  exp_ref,
  metadata_ref,
  vars = NULL,
  K = 100,
  verbose = FALSE,
  do_umap = TRUE,
  do_normalize = TRUE,
  vargenes_method = "vst",
  vargenes_groups = NULL,
  topn = 2000,
  tau = 0,
  theta = 2,
  save_uwot_path = NULL,
  d = 20,
  additional_genes = NULL,
  umap_min_dist = 0.1,
  seed = 111
)

Arguments

exp_ref

Reference gene expression (genes by cells)

metadata_ref

Reference cell metadata (cells by attributes)

vars

Reference variables to Harmonize over e.g. c('donor', 'technology')

K

Number of soft cluster centroids in model

verbose

Verbose output

do_umap

Perform UMAP visualization on harmonized reference embedding

do_normalize

Perform log(CP10K+1) normalization

vargenes_method

Variable gene selection method (either 'vst' or 'mvp')

vargenes_groups

Name of metadata column specifying groups for variable gene selection. If not NULL, calculate topn variable genes in each group separately, then pool

topn

Number of variable genes to subset by

tau

Tau parameter for Harmony step

theta

Theta parameter(s) for Harmony step

save_uwot_path

Absolute path to save the uwot model (used if do_umap is TRUE)

d

Number of PC dimensions

additional_genes

Any custom genes (e.g. marker genes) to include in addition to variable genes

umap_min_dist

umap parameter (see uwot documentation for details)

seed

Random seed

Value

Symphony reference object. Integrated embedding is stored in the $Z_corr slot. Other slots include cell-level metadata ($meta_data), variable genes means and standard deviations ($vargenes), loadings from PCA ($loadings), original PCA embedding ($Z_orig), reference compression terms ($cache), betas from Harmony integration ($betas), cosine normalized soft cluster centroids ($centroids), centroids in PC space ($centroids_pc), and optional umap coordinates ($umap$embedding).