Modality Alignment & Omics Layer Mixing in Joint Latent Space

Quantifies how well different single-cell omics layers mix in a unified latent embedding space without sacrificing biological clustering (Zhai et al., Genome Biology 2024; Luecken et al., Nat Methods 2022).

Usage

calc_modality_alignment(embedding, modalities, cell_types = NULL, k = 15)

Arguments

embedding

Joint coordinates matrix (cells x latent_dimensions).

modalities

Factor or character vector indicating the modality of each cell (e.g. "RNA" vs "ATAC").

cell_types

Optional factor or character vector of cell type labels.

k

Number of nearest neighbors for neighborhood connectivity calculation (default 15).

Value

A list containing modality_asw, modality_mixing_score, and mean_cross_modality_neighbor_frac.

Examples

data(example_multiomics, package = "scSimEval")
r_rna <- example_multiomics$ref_multi$rna
r_atac <- example_multiomics$ref_multi$atac
calc_modality_alignment(rbind(t(r_rna[seq_len(10),]), t(r_atac[seq_len(10),])), rep(c("RNA","ATAC"), each=ncol(r_rna)))
#> $modality_asw
#> [1] 0.3818172
#> 
#> $modality_mixing_score
#> [1] 0.6181828
#> 
#> $mean_cross_modality_neighbor_frac
#> [1] 0.0825
#>