1.数据准备

用10x数据做scVelo，输入数据分2部分:

• Seurat 分析后的数据: 聚类、基因、细胞信息。
• velocyto run 输出的loom文件，记录 spliced/unspliced 的RNA；

(1) Seurat 分析

假设你已经做好 Seurat 降维、分群、注释了。
velocyto 需要输入细胞的id。
一般多样本合并后的Seurat，需要按bam来源取子集，然后去掉cell id后缀再保存。参考一下几个文件:

> head(Cells(scObj_neu))
[1] "AAACCCAAGCCTCCAG-1_1" "AAACCCATCGGCATAT-1_1" "AAACGAATCCTGGGAC-1_1" "AAAGAACTCTTTACAC-1_1"
[5] "AAAGGATAGCGCAATG-1_1" "AAAGGATGTAGCTTTG-1_1"

tmp1=sapply(Cells(scObj_neu), function(x){
  (strsplit(x, "_")[[1]])[1]
}, USE.NAMES = F, simplify = T)
head(tmp1)
# [1] "AAACCCAAGCCTCCAG-1" "AAACCCATCGGCATAT-1"

writeLines(tmp1, paste0("scVelo/velocyto/WT2_barcodes.tsv") )

在 shell 中查看:

$ head ../scVelo/velocyto/WT2_barcodes.tsv
AAACCCACACACAGCC-1
AAACCCACAGACCAGA-1
AAACGAAGTCTTCGAA-1

包装好的函数如下: 半成品，因为函数第一行需要获取的是细胞和bam文件的编号，这两个信息每个人估计有自己的命名，根据自己情况改吧。

#' get cell barcode from sce, split by sample0
#'
#' @param sce 
#'
#' @return
#' @export
#'
#' @examples
getCid_list=function(sce){
  a1=FetchData(sce, vars=c("cell", "sample0"))
  #head(a1)
  strsplit(a1$cell, "_")[[1]]
  a1$cid=sapply(a1$cell, function(x){
    (strsplit(x, "_")[[1]])[1]
  }, USE.NAMES = F, simplify = T)
  #head(a1)
  #
  sapply(
    unique(a1$sample0),
    function(x){
      a1[which(a1$sample0==x),]$cid
    }, USE.NAMES = T
  )
}

# 测试1: 最终效果，就是一个cell id文件对应一个bam文件。
# APC是样本类型一样，但是来自不同的bam文件
sc_APC=subset(scObj_nue, origin=="APC")
sc_APC

# 默认按照 sample0 一列，也就是bam来源生成cell id 的list。
cid_APC=getCid_list(sc_APC)
str(cid_APC)
# 保存
lapply(names(cid_APC), function(x){
  message(x, "\t", length(cid_APC[[x]]), "\t", head(cid_APC[[x]]) )
  writeLines(cid_APC[[x]], paste0("./scVelo/velocyto/",x,"_barcodes.tsv") )
})


# 测试 2：最终效果，就是一个cell id文件对应一个bam文件。
writeLines(cid_WT[["WT1"]], "./202202newTag/scVelo/WT/WT1_barcodes.tsv")
writeLines(cid_WT[["WT2"]], "./202202newTag/scVelo/WT/WT2_barcodes.tsv")

(2) velocyto run 由bam输出loom文件(耗时)

http://velocyto.org/velocyto.py/tutorial/cli.html

Step 0: Constructing spliced and unspliced counts matrices
默认的 velocyto run10x 命令是使用的 filtered文件夹中的cell id，结果不理想。我还是喜欢原生的 velocyto run，更灵活，可控性更强。

需要准备几个输入文件

• cell id 文件，格式和可用代码见上文。
• 输出文件夹
• [可选] mask gtf 文件，一般从 UCSC 下载: expressed repeats annotation 猜测是染色体上过滤掉的区域。
• cell ranger 输出的bam文件
• 最后是 gtf 文件，和 cell range 用的gtf 基因组版本号要一致。

$ cd /home/wangjl/data/xx/scVelo/

$ velocyto run \
    -@ 100 --samtools-memory 1000 \
    -b WT/WT2_barcodes.tsv \
    -o WT/WT2 \
    -m ref/mouse_M25/mm10_rmsk.gtf \
    cellranger/WT2/outs/possorted_genome_bam.bam \
    ref/mouse_M25/gencode.vM25.annotation.gtf

#0:39-->02:01, 1.5h 只有600个细胞
WT/WT2/possorted_genome_bam_X3HXW.loom #22M

(3) 从 Seurat 输出数据

从Seurat对象导出细胞、基因、表达信息，供 scVelo 使用。

这是一个可用的函数，不用修改。

#' output Seurat obj info to a out dir/
#' 
#' V0.1
#'
#' @param sce Seurat obj
#' @param outputRoot a dir
#'
#' @return nothing
#' @export
#'
#' @examples
Seurat2scVelo=function(sce, outputRoot){
  message("output to:", outputRoot)
  # save metadata table:
  sce$barcode <- colnames(sce)
  sce$UMAP_1 <- sce@reductions$umap@cell.embeddings[,1]
  sce$UMAP_2 <- sce@reductions$umap@cell.embeddings[,2]
  #write.csv(data.frame(sce@meta.data)[,c("barcode","seurat_clusters", "UMAP_1","UMAP_2")], 
  write.csv(sce@meta.data, 
            file=paste0(outputRoot, 'metadata.csv'), quote=F, row.names=F)
  
  # write expression counts matrix
  library(Matrix)
  counts_matrix <- GetAssayData(sce, assay='RNA', slot='counts')
  writeMM(counts_matrix, file=paste0(outputRoot, 'counts.mtx'))
  
  # write dimesnionality reduction matrix, in this example case pca matrix
  write.csv(sce@reductions$pca@cell.embeddings, 
            file=paste0(outputRoot, 'pca.csv'), quote=F, row.names=F)
  
  # write gene names
  write.table(
    data.frame('gene'=rownames(counts_matrix)),
    file=paste0(outputRoot, 'gene_names.csv'),
    quote=F,row.names=F,col.names=F
  )
  return(invisible(NULL))
}

# 测试:
Seurat2scVelo(sc_WT, "202202newTag/scVelo/Seurat_dataset/WT/")

2. 使用 Python 预处理数据

使用 Jupyter notebook.

(1) 读取 Seurat 导出的数据

import os
os.chdir("/home/wangjl/data/neu/scRNA/202202newTag/scVelo/")
os.getcwd()


import scanpy as sc
import anndata 
from scipy import io
from scipy.sparse import coo_matrix, csr_matrix
import numpy as np
import os
import pandas as pd

type="mergedWT"

# pip install anndata --force-reinstall
#anndata  0.7.8

#(1) load sparse matrix:
X = io.mmread(f"Seurat_dataset/{type}/counts.mtx")

# create anndata object
adata = anndata.AnnData(
    X=X.transpose().tocsr()
)


#(2) load cell metadata:
cell_meta = pd.read_csv(f"Seurat_dataset/{type}/metadata.csv")

# load gene names:
with open(f"Seurat_dataset/{type}/gene_names.csv", 'r') as f:
    gene_names = f.read().splitlines()

# set anndata observations and index obs by barcodes, var by gene names
adata.obs = cell_meta
adata.obs.index = adata.obs['barcode']

adata.var.index = gene_names

#(3) load dimensional reduction:
pca = pd.read_csv(f"Seurat_dataset/{type}/pca.csv")
pca.index = adata.obs.index

# set pca and umap
adata.obsm['X_pca'] = pca.to_numpy()
adata.obsm['X_umap'] = np.vstack((adata.obs['UMAP_1'].to_numpy(), adata.obs['UMAP_2'].to_numpy())).T

# plot a UMAP colored by sampleID to test:
sc.pl.umap(adata, color='seurat_clusters', frameon=False, save=True)

(2) 读取 loom 数据(unsplied/spliced data)

import scvelo as scv
import scanpy as sc
#import cellrank as cr
import numpy as np
import pandas as pd
import anndata as ad

import matplotlib.pyplot as plt

scv.settings.verbosity = 3
scv.settings.set_figure_params('scvelo', facecolor='white', dpi=100, 
                               fontsize=6, color_map='viridis',
                               frameon=False)

#cr.settings.verbosity = 2
#adata = sc.read_h5ad('my_data.h5ad')
#adata=sc.read_loom(f'out/{type}.my_data.loom')

#adata.obsm['X_pca'] = pca.to_numpy()
#adata.obsm['X_umap'] = np.vstack((adata.obs['UMAP_1'].to_numpy(), adata.obs['UMAP_2'].to_numpy())).T

#adata

# 这里展示了多个样本的处理方式。

# load loom files for spliced/unspliced matrices for each sample:
sample_loom_1="./velocyto/WT1/possorted_genome_bam_HA9DP.loom"
ldata1 = scv.read( sample_loom_1, cache=True)
ldata1

sample_loom_2="./velocyto/WT2/possorted_genome_bam_SOJRI.loom"
ldata2 = scv.read( sample_loom_2, cache=True)
ldata2

(3) merge loom data

ldata1.obs.index[0:2]

加上cell id后缀，和Seurat 中的一致。

# WT_1 -1_1
barcodes = [bc.split(':')[1] for bc in ldata1.obs.index.tolist()]
barcodes = [bc[0:len(bc)-1] + '-1_1' for bc in barcodes]
ldata1.obs.index = barcodes
ldata1.obs.index[0:5]

# WT2 -1_2
barcodes = [bc.split(':')[1] for bc in ldata2.obs.index.tolist()]
barcodes = [bc[0:len(bc)-1] + '-1_2' for bc in barcodes]
ldata2.obs.index = barcodes
ldata2.obs.index[0:5]

ldata1.var.head()

ldata1.var_names_make_unique()
ldata2.var_names_make_unique()

合并2个loom文件

# merge
ldata = ldata1.concatenate([ldata2])
ldata

和Seurat数据合并

# merge matrices into the original adata object
adata = scv.utils.merge(adata, ldata)
adata

保存数据

adata.write_h5ad(f'data/{type}.adata_ldata.h5ad')

(4) scVelo 预处理

adata = sc.read(f'data/{type}.adata_ldata.h5ad')
adata

#scv.pp.filter_and_normalize(adata, min_shared_counts=5, min_shared_cells=3, log=True)
scv.pp.filter_and_normalize(adata)

可选步骤:

## clean some genes
import re
flag = [not bool(re.match('^Rp[ls]', i)) for i in adata.var_names]
adata = adata[:,flag]
#
adata = adata[:,adata.var_names != "Malat1"]
adata

3.scVelo

scv.pp.moments(adata, n_neighbors=30, n_pcs=30)

# this step will take a long while
import gc
gc.collect()
#
temp_pre= f"{type}_nue.in_process2"
if False==os.path.exists(f"{temp_pre}.velo.gz.h5ad"):
    scv.tl.recover_dynamics(adata, var_names='all', n_jobs=64)
    scv.tl.velocity(adata, mode='dynamical')
    adata.write(f"{temp_pre}.velo.gz.h5ad", compression='gzip')
    print(">>Write to file")
else:
    adata = sc.read(f"{temp_pre}.velo.gz.h5ad", compression='gzip', ext="h5ad")
    print(">>read from file")

scv.tl.velocity_graph(adata)
scv.tl.velocity_embedding(adata, basis="umap")

(2) 可视化

add color

## add colSet
adata.obs["cluster_names"] = adata.obs["cluster_names"].astype('category')

color = pd.read_csv(f"../merge/data/mergeR3.neutrophil.color.txt", #compression="gzip", 
            sep=" ", header=0, index_col=0)
#color.index=color.index.astype("str")
color.index=color["cluster_names"]

# 对数据 按分类因子 排序
adata.obs["cluster_names"].cat.reorder_categories(color["cluster_names"], inplace=True)

# 获取颜色列
color_used = list(color.loc[adata.obs["cluster_names"].cat.categories,"color"])
adata.uns['cluster_names_colors'] = color_used
adata

embedding

scv.pl.velocity_embedding(adata, basis = 'umap', title="WT",
                          save=f"{type}.embedding.pdf",
                          color="seurat_clusters")

grid

scv.settings.set_figure_params('scvelo', dpi=300, dpi_save=300)
fig, ax = plt.subplots()
ax.set_aspect(1)

scv.pl.velocity_embedding_grid(adata, basis='umap',color='cluster_names', title='WT',
                               arrow_size=1, arrow_length=2, arrow_color="#D2691E",
                               alpha=0.1,
                               #density=0.9,
                               legend_loc='right margin',legend_fontsize=5,
                               show=True,
                               save=f"{type}.grid.pdf", #figsize=(10,10),
                               xlim=[-10,10],ylim=[-10,10], ax=ax)

# 默认
scv.settings.set_figure_params('scvelo', dpi=300, dpi_save=300)
fig, ax = plt.subplots()
ax.set_aspect(1)
scv.pl.velocity_embedding_grid(adata, basis='umap',color='cluster_names', title='WT',
                               arrow_size=1, arrow_length=2, arrow_color="#D2691E",
                               alpha=0.1,
                               #density=0.9,
                               legend_loc='right margin',legend_fontsize=5,
                               show=True,
                               save=f"{type}.grid.pdf", #figsize=(10,10),
                               xlim=[-10,10],ylim=[-10,10], ax=ax)

stream

%matplotlib inline
import matplotlib.pyplot as plt
scv.settings.set_figure_params('scvelo', dpi=300, dpi_save=300)
#fig, ax = plt.subplots()
#ax.set_aspect(1)
scv.pl.velocity_embedding_stream(adata, basis='umap',color='cluster_names', title='WT',
                               #arrow_size=1, ##arrow_length=2, 
                               #arrow_color="#D2691E",
                               #alpha=0.01, density=0.9,
                               legend_loc='right margin',legend_fontsize=5,
                               show=True,
                               save=f"{type}.stream.pdf")
                                #, #figsize=(10,10),
                               #xlim=[-10,10],ylim=[-10,10], 
                               #  ax=ax)

scv.pl.velocity_embedding_stream(adata, basis="umap", title='WT',
                                 save=f"{type}.stream2.pdf",
                                 color="cluster_names")

for each sample

# for each cancerType (only T)
for i in ['WT1','WT2']:
    flag = [j == i for j in adata.obs.loc[:,'sample0']]
    adata_sub = adata[flag,]
    #flag2 = [j == "T" for j in adata_sub.obs.loc[:,'loc']]
    #adata_sub = adata_sub[flag2,]
    #
    scv.settings.set_figure_params('scvelo', dpi=300, dpi_save=300)
    fig, ax = plt.subplots()
    ax.set_aspect(1)
    scv.pl.velocity_embedding_grid(adata_sub, basis='umap',color='cluster_names', title=i,
                               arrow_size=1, arrow_length=2, arrow_color="#D2691E", alpha=0.1,
                               legend_loc='right margin',legend_fontsize=5,
                               density=0.8,
                               save=f"{type}_Sup.{i}.grid.pdf",# figsize=(10,9),
                               xlim=[-10,10],ylim=[-10,10], ax=ax)

latent time

scv.tl.latent_time(adata)
scv.pl.scatter(adata, color='latent_time', 
               save=f"{type}.latent_time.pdf",# figsize=(10,9),
               color_map='gnuplot', size=80)

heatmap

top_genes = adata.var['fit_likelihood'].sort_values(ascending=False).index[:300] 
scv.pl.heatmap(adata, var_names=top_genes, sortby='latent_time', col_color='cluster_names', 
               save=f"{type}.heatmap.pdf",# figsize=(10,9),
               n_convolve=100)

Top-likelihood genes

# 查看top15的基因 
#top_genes = adata.var['fit_likelihood'].sort_values(ascending=False).index 
scv.pl.scatter(adata, basis=top_genes[:15], ncols=5, 
               save=f"{type}.Driver_genes.pdf",# figsize=(10,9),
               frameon=False)

更多可视化，参考

• https://scvelo.readthedocs.io/
• https://scvelo.readthedocs.io/VelocityBasics/
• https://scvelo.readthedocs.io/DynamicalModeling/
• https://scvelo.readthedocs.io/DifferentialKinetics/

ref

https://www.jianshu.com/p/fb1cf5806912
https://www.jianshu.com/p/bfff8a4cf611