Convert Seurat to Scanpy h5ad

1
2
3

# in R console
## install myfork, which compatible with latest anndata 
remotes::install_github("zqfang/MuDataSeurat", force = T)

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21

# (optional) step 1: Slim down a Seurat object. So you get raw counts, lognorm counts
seu = DietSeurat(
  srt,
  counts = TRUE, # so, raw counts save to adata.layers['counts']
  data = TRUE, # so, log1p counts save to adata.X when scale.data = False, else adata.layers['data']
  scale.data = FALSE, # set to false
  features = rownames(srt), # export all genes, not just top highly variable genes
  assays = "RNA",
  dimreducs = c("pca","umap"),
  graphs = c("RNA_nn", "RNA_snn"), # to RNA_nn -> distances, RNA_snn -> connectivities
  misc = TRUE
)


# .X, .layers['counts']. .raw.X
# Assumptions:
#   1. counts/data/scale.data -> X
#   3. counts & data -> layers['counts'], X
#   2. data & scale.data -> layers['data'], X
#   4. counts & scale.data -> layers['counts'], X
#   5. counts & data & scale.data -> layers['counts'], layers['data'], X

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18

# for seurat v5, need to JoinLayer first
DefaultAssay(srt) = "RNA"
srt <- JoinLayers(srt)
# single modality
MuDataSeurat::WriteH5AD(srt, "srt.h5ad", assay="RNA")
# multi modality, ATAC+RNA, CITE-seq 
MuDataSeurat::WriteH5MU(srt, "srt.h5mu")

## data mapping
## GEX
# counts -> adata.layers['counts']
# data -> adata.layers['data'] # log1p data
# scale.data -> adata.X
# adata.raw.X set to None
## Graph
# RNA_nn -> adata.obsp['nn'] # this is knn adjmat. the old .uns['neighbors']['distance']
# RNA_snn -> adata.obsp['snn] # umap graph. the old .uns['neighbors']['connectivies']
## 

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12

library(sceasy)
library(reticulate)
use_condaenv('EnvironmentName')
loompy <- reticulate::import('loompy')

# seurat to anndata
sceasy::convertFormat(seurat_object, from="seurat", to="anndata",
                       outFile='filename.h5ad')

# anndata to seurat
sceasy::convertFormat(h5ad_file, from="anndata", to="seurat",
                       outFile='filename.rds')

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

library(Seurat)
library(SeuratDisk)

# step 1: Slim down a Seurat object. So you get raw counts, lognorm counts

seu = DietSeurat(
  srt,
  counts = TRUE, # so, raw counts save to adata.raw.X 
  data = TRUE, # so, log1p counts save to adata.X
  scale.data = FALSE, # set to false, or else will save to adata.X
  features = rownames(srt), # export all genes, not just top highly variable genes
  assays = "RNA",
  dimreducs = c("pca","umap"),
  graphs = c("RNA_nn", "RNA_snn"), # to RNA_nn -> distances, RNA_snn -> connectivities
  misc = TRUE
)

# step 2: factor to character, or else your factor will be number in adata 
i <- sapply(seu@meta.data, is.factor)
seu@meta.data[i] <- lapply(seu@meta.data[i], as.character)

# step 3: convert 
SaveH5Seurat(seu, filename = "srt.h5seurat", overwrite = TRUE)
Convert("srt.h5seurat", "srt.h5ad", assay="RNA", overwrite = TRUE)

1
2
3
4
5
6
7
8

# load h5ad
import scanpy as sc
adata = sc.read_h5ad("srt.h5ad")
# save counts to layers
adata.layers['counts'] = adata.raw.X.copy()
adata.layers['log1p'] = data.X.copy()

# you need to scale data for downsream tasks if needed.

1
2
3
4
5
6
7

pbmc.loom <- as.loom(seurat_object, filename = "../output/pbmc.loom", verbose = FALSE)
pbmc.loom$close_all() # alway close when done 

write.csv(Cells(seurat_object), file = "cellID_obs.csv", row.names = FALSE)
write.csv(Embeddings(seurat_object, reduction = "umap"), file = "cell_embeddings.csv")
write.csv(seurat_object@meta.data$seurat_clusters, file = "clusters.csv")

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13

import scanpy as sc
adata = sc.read_loom("../output/pbmc.loom")

sample_obs = pd.read_csv("cellID_obs.csv")
umap_cord = pd.read_csv("cell_embeddings.csv")
cell_clusters = pd.read_csv("clusters_obs.csv")


# now add metadata to the adata 
# e.g.
adata.obsm['X_umap'] = umap_ordered.values # cellID should be matched first
adata.uns['Cluster_colors'] = ...
...

1
2
3

immune <- Connect(filename = "../data/mmune_cells.loom", mode = "r")
seurat <- as.Seurat(immune)
immune$close_all() # alway close when done 

1
2
3
4
5
6
7

# user defined R installation
import os
# path to your libR.so, only Seurat v3.0.2 works! 
# create a conda R env for seurat 3.0.2 first
os.environ['R_HOME'] = '/home/fangzq/miniconda/envs/seurat/lib/R' 
# path depends on where you installed Python.
os.environ['R_USER'] = '/home/fangzq/miniconda/lib/python3.7/site-packages/rpy2' 

1
2

import scanpy as sc
import glob

1
2
3
4
5

import anndata2ri
from rpy2.robjects import r
from rpy2.robjects.conversion import localconverter
# activate rpy2 env
anndata2ri.activate()

1

robjs = glob.glob("data/*Robj")

 1
 2
 3
 4
 5
 6
 7
 8
 9
10

r('library(Seurat)')
for robj in robjs:
    r(f'x<-load("{robj}")')
    r('y=get(x)')
    r('rm(x)')
    r('DefaultAssay(y) <- "RNA"') # get raw count matrix to save
    # seurat2 object
    # adata = r('as.SingleCellExperiment(UpdateSeuratObject(y))')
    adata = r('as.SingleCellExperiment(y)')
    adata.write_h5ad(filename=robj.replace("Robj","h5ad"))