We’ve found that by using anndata for R, interacting with other anndata-based Python packages becomes super easy!
Let’s use a 10x dataset from the 10x genomics website. You can download it to an anndata object with scanpy as follows:
library(anndata)
library(reticulate)
<- import("scanpy")
sc
<- "https://cf.10xgenomics.com/samples/cell-exp/6.0.0/SC3_v3_NextGem_DI_CellPlex_CSP_DTC_Sorted_30K_Squamous_Cell_Carcinoma/SC3_v3_NextGem_DI_CellPlex_CSP_DTC_Sorted_30K_Squamous_Cell_Carcinoma_count_sample_feature_bc_matrix.h5"
url <- sc$read_10x_h5("dataset.h5", backup_url = url)
ad
ad#> AnnData object with n_obs × n_vars = 5377 × 36601
#> var: 'gene_ids', 'feature_types', 'genome'
The resuling dataset is a wrapper for the Python class but behaves very much like an R object:
1:5, 3:5]
ad[#> View of AnnData object with n_obs × n_vars = 5 × 3
#> var: 'gene_ids', 'feature_types', 'genome'
dim(ad)
#> [1] 5377 36601
But you can still call scanpy functions on it, for example to perform preprocessing.
$pp$filter_cells(ad, min_genes = 200)
sc$pp$filter_genes(ad, min_cells = 3)
sc$pp$normalize_per_cell(ad)
sc$pp$log1p(ad) sc
You can seamlessly switch back to using your dataset with other R functions, for example by calculating the rowMeans of the expression matrix.
library(Matrix)
rowMeans(ad$X[1:10, ])
#> AAACCCAAGCGCGTTC-1 AAACCCAAGGCAATGC-1 AAACCCAGTATCTTCT-1 AAACCCAGTGACAACG-1
#> 0.05451418 0.13627126 0.12637224 0.13958617
#> AAACCCAGTTGAATCC-1 AAACCCATCGGCTTGG-1 AAACGAAAGAGAGCCT-1 AAACGAAAGCTTAAGA-1
#> 0.05979424 0.11365747 0.05011727 0.14347849
#> AAACGAAAGGCACGAT-1 AAACGAAAGGTAGCCA-1
#> 0.12979302 0.12366312