How to convert between Seurat/SingleCellExperiment object and Scanpy object/AnnData using basic packages

1. From Scanpy object to Seurat object

The basic idea is saving to and reading from the disk.

Firstly, we need to write the data in Python into the disk:

from scipy import io
save_dir='/home/daimin/result/single-cell/hcm/analysis'
io.mmwrite(save_dir+'/counts.mtx',adata.layers['raw'])

Save the meta data for both cells and genes:

cell_meta=adata.obs.copy()
cell_meta['Barcode']=cell_meta.index
cell_meta['UMAP1']=adata.obsm['X_umap'][:,0]
cell_meta['UMAP2']=adata.obsm['X_umap'][:,1]
gene_meta=adata.var.copy()
gene_meta['GeneName']=gene_meta.index
cell_meta.to_csv(save_dir+'/counts_cellMeta.csv',index=None)
gene_meta.to_csv(save_dir+'/counts_geneMeta.csv',index=None)

Next, we load the data from the disk into the R environment:

library(Matrix)
save_dir<-'/home/daimin/result/single-cell/IN/Satb2_ko_CGE'
counts<-readMM(paste0(save_dir,'/counts.mtx'))
dim(counts)
cellMeta<-read.csv(paste0(save_dir,'/counts_cellMeta.csv'))
head(cellMeta)
geneMeta<-read.csv(paste0(save_dir,'/counts_geneMeta.csv'))
dim(geneMeta)
head(geneMeta)
### Set the rownames and colnames
rownames(counts)<-cellMeta$Barcode
colnames(counts)<-geneMeta$GeneName

Then, for example, you can construct a Seurat object:

seo <- CreateSeuratObject(counts = t(counts), project = "min", min.cells = 3, min.features = 200)

2. How to load the sparse matrix in R into Python

For example, your data looks like this:

4 x 4 sparse Matrix of class "dgCMatrix"
              SM-GE4R1_S041_E1-50 SM-GE4R1_S042_E1-50 SM-GE4R1_S043_E1-50 SM-GE4R1_S044_E1-50
0610005C13Rik              .                   .                  .                   .      
0610006L08Rik              .                   .                  .                   .      
0610007P14Rik            137.8368            138.3161           143.01993            70.96005
0610009B22Rik            109.6429            124.8218            73.06453            11.82667

and now save the data into the disk using Matrix’s wirteMM function:

writeMM(cpm,'/home/daimin/data/single_cell/mouse_brain_vmh/SMART-seq_VMH_cpm.mtx')

The saved mtx file looks like this:

%%MatrixMarket matrix coordinate real general
30862 4574 41234632
3 1 137.8368387747558
4 1 109.64293993446489
5 1 9.39796628009699
6 1 6.265310853397991
7 1 37.59186512038795
9 1 46.98983140048494
10 1 46.98983140048494
11 1 126.8725447813093

Then, save the barcodes:

barcodes<-data.frame(colnames(cpm))
colnames(barcodes)<-'Barcode'
write.csv(barcodes, '/home/daimin/data/single_cell/mouse_brain_vmh/SMART-seq_VMH_cpm_barcodes.csv',
          quote = FALSE,row.names = FALSE)

and the data looks like this:

Barcode
SM-GE4R1_S041_E1-50
SM-GE4R1_S042_E1-50
SM-GE4R1_S043_E1-50
SM-GE4R1_S044_E1-50
SM-GE4R1_S045_E1-50
SM-GE4R1_S046_E1-50
SM-GE4R1_S047_E1-50
SM-GE4R1_S048_E1-50
SM-GE4R1_S049_E1-50

Save the gene names:

genes<-data.frame(rownames(cpm))
colnames(genes)<-'Gene'
write.csv(genes, '/home/daimin/data/single_cell/mouse_brain_vmh/SMART-seq_VMH_cpm_genes.csv',
          quote = FALSE,row.names = FALSE)

and the file looks like:

Gene
0610005C13Rik
0610006L08Rik
0610007P14Rik
0610009B22Rik
0610009E02Rik
0610009L18Rik
0610009O20Rik
0610010B08Rik
0610010F05Rik
0610010K14Rik
0610011F06Rik
0610012G03Rik
0610025J13Rik
0610030E20Rik

Next, we load the data in Python:

import scanpy as sc
import pandas as pd
from scipy import io
counts=io.mmread('/home/daimin/data/single_cell/mouse_brain_vmh/SMART-seq_VMH_cpm.mtx')
from scipy import sparse
counts=sparse.csr_matrix(counts)
barcodes=pd.read_csv('/home/daimin/data/single_cell/mouse_brain_vmh/SMART-seq_VMH_cpm_barcodes.csv')
genes=pd.read_csv('/home/daimin/data/single_cell/mouse_brain_vmh/SMART-seq_VMH_cpm_genes.csv')
adata=sc.AnnData(counts.T)
adata.obs_names=barcodes['Barcode'].values
adata.var_names=genes['Gene'].values

and now check the adata:

adata
# AnnData object with n_obs × n_vars = 4574 × 30862