SpotClean: Decontaminate spot swapping effect in spatial transcriptomics...

In zijianni/SpotClean: SpotClean adjusts for spot swapping in spatial transcriptomics data

Decontaminate spot swapping effect in spatial transcriptomics data

Description

This is the main function implementing the SpotClean method for decontaminating spot swapping effect in spatial transcriptomics data.

Usage

spotclean(slide_obj, ...)

## S3 method for class 'SummarizedExperiment'
spotclean(
  slide_obj,
  gene_keep = NULL,
  maxit = 30,
  tol = 1,
  candidate_radius = 5 * seq_len(6),
  kernel = "gaussian",
  verbose = TRUE,
  ...
)

## S3 method for class 'SpatialExperiment'
spotclean(
  slide_obj,
  gene_keep = NULL,
  gene_cutoff = 0.1,
  maxit = 30,
  tol = 1,
  candidate_radius = 5 * seq_len(6),
  kernel = "gaussian",
  verbose = TRUE,
  ...
)

Arguments

slide_obj	A slide object created or inherited from createSlide(), or a SpatialExperiment object created from SpatialExperiment::read10xVisium().
...	Arguments passed to other methods
gene_keep	(vector of chr) Gene names to keep for decontamination. We recommend not decontaminating lowly expressed and lowly variable genes in order to save computation time. Even if user include them, their decontaminated expressions will not change too much from raw expressions. When setting to NULL, keepHighGene() will be automatically called to filter out lowly expressed and lowly variable genes before decontamination. Default: NULL.
maxit	(int) Maximum iteration for EM parameter updates. Default: 30.
tol	(num) Tolerance to define convergence in EM parameter updates. When the element-wise maximum difference between current and updated parameter matrix is less than tol, parameters are considered converged. Default: 1
candidate_radius	(vector of num) Candidate contamination radius. A series of radius to try when estimating contamination parameters. Default: c(5, 10, 15, 20, 25, 30)
kernel	(chr): name of kernel to use to model local contamination. Supports "gaussian", "linear", "laplace", "cauchy". Default: "gaussian".
verbose	(logical) Whether print progress information. Default: TRUE
gene_cutoff	(num) Filter out genes with average expressions among tissue spots below or equal to this cutoff. Only applies to SpatialExperiment object. Default: 0.1.

Details

Briefly, the contamination level for the slide is estimated based on the total counts of all spots. UMI counts travelling around the slide are assumed to follow Poisson distributions and modeled by a mixture of Gaussian (proximal) and uniform (distal) kernels. The underlying uncontaminated gene expressions are estimated by EM algorithm to maximize the data likelihood. Detailed derivation can be found in our manuscript.

Value

For slide object created from createSlide(), returns a slide object where the decontaminated expression matrix is in the "decont" assay slot and the contamination statistics are in metadata slots. Contamination statistics include ambient RNA contamination (ARC) score, bleeding rate, distal rate, contamination radius, contamination kernel weight matrix, log-likelihood value in each iteration, estimated proportion of contamination in each tissue spot in observed data. Since decontaminated and raw data have different number of columns, they can not be stored in a single object.

For SpatialExperiment object created from SpatialExperiment::read10xVisium(), returns a SpatialExperiment object where the decontaminated expression matrix is in the "decont" assay slot and the contamination statistics are in metadata slots. Raw expression matrix is also stored in the "counts" assay slot. Genes are filtered based on gene_cutoff.

Examples

data(mbrain_raw)
spatial_dir <- system.file(file.path("extdata",
                                     "V1_Adult_Mouse_Brain_spatial"),
                           package = "SpotClean")
mbrain_slide_info <- read10xSlide(tissue_csv_file=file.path(spatial_dir,
                                       "tissue_positions_list.csv"),
             tissue_img_file = file.path(spatial_dir,
                                       "tissue_lowres_image.png"),
             scale_factor_file = file.path(spatial_dir,
                                       "scalefactors_json.json"))
mbrain_obj <- createSlide(mbrain_raw,
                          mbrain_slide_info)

mbrain_decont_obj <- spotclean(mbrain_obj, tol=10, candidate_radius=20)
mbrain_decont_obj

zijianni/SpotClean documentation built on Nov. 15, 2023, 12:53 a.m.