calcNhoodSim: Compare neighbourhoods across species

In dkeitley/scrabbitr: Single-cell RNA-seq analysis of rabbit gastrulation

Compare neighbourhoods across species

Description

Runs the neighbourhood comparison pipeline

Usage

calcNhoodSim(
  r_milo,
  m_milo,
  orthologs,
  sim_preprocessing = "gene_spec",
  sim_measure = "pearson",
  hvg_join_type = "intersection",
  r_exclude = NULL,
  m_exclude = NULL,
  export_dir = NULL,
  verbose = TRUE,
  ...
)

Arguments

r_milo	The rabbit Milo object
m_milo	The mouse Milo object
orthologs	A DataFrame of rabbit and mouse one-to-one orthologs.
sim_preprocessing	Used to specify whether to compute the gene specificity before computing similiarities between neighbourhoods. Can be either "gene_spec" or "none".
sim_measure	The similarity measure to compare neighbourhoods. Can be one of "pearson", "kendall" or "spearman".
hvg_join_type	Specifies how to combine gene features from the two species. Can be either "intersection" or "union".
r_exclude	A vector of rabbit genes to exclude from feature selection.
m_exclude	A vector of mouse genes to exclude from feature selection.
export_dir	A string path indicating where to export output files.
verbose	A logical scalar indicating whether progress updates should be printed to screen.
...	Additional arguments to pass to specific methods. See ?selectNhoodFeatures.

Details

This function implements the neighbourhood comparison pipeline described in Ton et al. (2022).

The pipeline consists of two main steps - selecting features and computing neighbourhood similarities.

Feature selection

Fistly, to remove uninformative genes from driving similarities between species, a subset of genes are chosen independently for each species by calling selectNhoodFeatures. By default this uses scran::getScranHVGs, however, it's also possible to specify a set of predefined features by passing a list of genes to the hvg_selection argument.

These genes are then filtered to exclude genes that are listed in r_exclude and m_exclude. Only genes that are one-to-one orthologs, as specified in orthologs are retained.

The features from each species are then combined according to hvg_join_type. This provides a common gene set with which to compute similarities between neighbourhoods.

Computing neighbourhood similarities

Using these selected features, a mean expression profile is then computed for each neighbourhood using calcNhoodMean. These expression values are extracted using the logcounts assay of the rabbit and mouse Milo object which must represent normalised logcounts.

Prior to computing the similarity between neighbourhoods an additional normalisation step can be performed using the sim_preprocessing = "gene_spec" parameter option. Specifically, the gene specificity (s^i_g) is computed for each neighbourhood, given by

s^i_g = N*g_i / (g_1 + g_2 + ... + g_N)

where g_x represents the mean expression of gene g in neighbourhood x = 1,...,N. This can be used to account for differences in absolute values between datasets.

Following this, the similarity between neighbourhoods is computed using the correlation measure specified by sim_measure.

The matrix of rabbit vs mouse neighbourhood similarity scores, as well as mean expression values (or gene specificity values if sim_preprocessing = "gene_spcec") are exported to the directory specified by export_dir.

Value

A named list is returned containing:

r_vals:

A DataFrame of mean expression values (or gene specificity values) for each rabbit gene across all rabbit neighbourhoods.

m_vals:

A DataFrame of mean expression values (or gene specificity values) for each mouse gene across all mouse neighbourhoods.

nhood_sim:

An all vs all matrix of similarities between rabbit (rows) and mouse (columns) neighbourhoods.

Author(s)

Daniel Keitley

References

Ton M.L, Keitley D et al. 2022 Rabbit Development as a Model for Single Cell Comparative Genomics Manuscript in submission.

See Also

selectNhoodFeatures, calcNhoodMean, calcGeneSpec and exportNhoodSim for more info on specific steps of the pipeline.

See an example usage here.

dkeitley/scrabbitr documentation built on Feb. 13, 2023, 4:26 p.m.