Python单细胞复现2022||03-多样本整合分析

单个样本分析接上一篇：Python图文复现2022||02-数据介绍与下载，结合视频观看效果更佳~

视频相关代码如下：

• 视频如下：https://www.youtube.com/watch?v=uvyG9yLuNSE
• 代码：https://github.com/mousepixels/sanbomics_scripts
• 主代码：https://github.com/mousepixels/sanbomics_scripts/blob/main/single_cell_analysis_complete_class.ipynb

读取数据

先定义一个函数，批量运行多个样本：这里一定要注意缩进问题

def pp(csv_path):
    adata = sc.read_csv(csv_path).T
    sc.pp.filter_genes(adata, min_cells = 10)
    sc.pp.highly_variable_genes(adata, n_top_genes = 2000, subset = True, flavor = 'seurat_v3')
    scvi.model.SCVI.setup_anndata(adata)
    vae = scvi.model.SCVI(adata)
    vae.train()
    solo = scvi.external.SOLO.from_scvi_model(vae)
    solo.train()
    df = solo.predict()
    df['prediction'] = solo.predict(soft = False)
    df.index = df.index.map(lambda x: x[:-2])
    df['dif'] = df.doublet - df.singlet
    doublets = df[(df.prediction == 'doublet') & (df.dif > 1)]
    adata = sc.read_csv(csv_path).T
    adata.obs['Sample'] = csv_path.split('_')[2] #'raw_counts/GSM5226574_C51ctr_raw_counts.csv'
    adata.obs['doublet'] = adata.obs.index.isin(doublets.index)
    adata = adata[~adata.obs.doublet]
    sc.pp.filter_cells(adata, min_genes=200) #get rid of cells with fewer than 200 genes
    #sc.pp.filter_genes(adata, min_cells=3) #get rid of genes that are found in fewer than 3 cells
    adata.var['mt'] = adata.var_names.str.startswith('mt-')  # annotate the group of mitochondrial genes as 'mt'
    adata.var['ribo'] = adata.var_names.isin(ribo_genes[0].values)
    sc.pp.calculate_qc_metrics(adata, qc_vars=['mt', 'ribo'], percent_top=None, log1p=False, inplace=True)
    upper_lim = np.quantile(adata.obs.n_genes_by_counts.values, .98)
    adata = adata[adata.obs.n_genes_by_counts < upper_lim]
    adata = adata[adata.obs.pct_counts_mt < 20]
    adata = adata[adata.obs.pct_counts_ribo < 2]
    return adata

这个过程比较久，会依次读取GSE171524数据集中的26例样本，并进行上面的函数里面定义的分析。

这个过程中就可以跑去听听视频了。

import os
# 注意dir改成自己的路径
dir = '/path/data/GSE171524/'
out = []
for file in os.listdir(dir+'raw_counts/'):
    out.append(pp(dir + 'raw_counts/' + file))

将多个样本连接合并在一起，合并后共有105264个细胞：

adata = sc.concat(out)
adata

AnnData object with n_obs × n_vars = 105264 × 29236
    obs: 'Sample', 'doublet', 'n_genes', 'n_genes_by_counts', 'total_counts', 'total_counts_mt', 'pct_counts_mt', 'total_counts_ribo', 'pct_counts_ribo'
    var: 'n_cells'

过滤并保存：

sc.pp.filter_genes(adata, min_cells = 10)

from scipy.sparse import csr_matrix
adata.X = csr_matrix(adata.X)
adata.X
adata.write_h5ad(dir+'combined.h5ad')

预处理

先读取上次保存的数据：105264的细胞

import scanpy as sc
import scvi
import seaborn as sns
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

# 注意dir改成自己的路径
dir = '/path/data/GSE171524/'
adata = sc.read_h5ad(dir+'combined.h5ad')
adata

AnnData object with n_obs × n_vars = 105264 × 29236
    obs: 'Sample', 'doublet', 'n_genes', 'n_genes_by_counts', 'total_counts', 'total_counts_mt', 'pct_counts_mt', 'total_counts_ribo', 'pct_counts_ribo'
    var: 'n_cells'

每个样本中的各种指标统计：

adata.obs.groupby('Sample').count()

共26个样本：

image-20221001224829602.png

低表达过滤以及数据标准化

sc.pp.filter_genes(adata, min_cells = 100)
adata.layers['counts'] = adata.X.copy()
sc.pp.normalize_total(adata, target_sum = 1e4)
sc.pp.log1p(adata)
adata.raw = adata

# 查看每个细胞的指标
adata.obs.head()

结果图：

image-20221001225018337.png

去Sample间的批次以及聚类

这个过程运行时间也会有丢丢长

scvi.model.SCVI.setup_anndata(adata, layer = "counts", categorical_covariate_keys=["Sample"], continuous_covariate_keys=['pct_counts_mt', 'total_counts', 'pct_counts_ribo'])

model = scvi.model.SCVI(adata)

#may take a while without GPU
model.train()

adata.obsm['X_scVI'] = model.get_latent_representation()
adata.layers['scvi_normalized'] = model.get_normalized_expression(library_size = 1e4)
sc.pp.neighbors(adata, use_rep = 'X_scVI')
sc.tl.umap(adata)
sc.tl.leiden(adata, resolution = 0.5)
sc.pl.umap(adata, color = ['leiden', 'Sample'], frameon = False)
plt.savefig(dir+"01-intergration_umap.png")

结果图：

image-20221002000706224.png

保存数据：

adata.write_h5ad(dir + 'integrated.h5ad')

差异表达分析

使用没有矫正的数据做差异表达分析

# 更改聚类数
sc.tl.leiden(adata, resolution = 1)
sc.tl.rank_genes_groups(adata, 'leiden')

# 可视化
sc.pl.rank_genes_groups(adata, n_genes=20, sharey=False)
plt.savefig(dir+"02-intergration_rank_genes_groups.png", dpi=300)

部分cluster的top基因：

image-20221002204205291.png

markers = sc.get.rank_genes_groups_df(adata, None)
markers = markers[(markers.pvals_adj < 0.05) & (markers.logfoldchanges > .5)]
markers

# 保存
markers.to_csv(dir+'markers.csv')

差异结果：

image-20221002002640069.png

使用模型标准化后的值做差异表达分析：

markers_scvi = model.differential_expression(groupby = 'leiden')

# 设定阈值
markers_scvi = markers_scvi[(markers_scvi['is_de_fdr_0.05']) & (markers_scvi.lfc_mean > .5)]
markers_scvi

# 保存
markers_scvi.to_csv(dir+'scvi_markers.csv')

结果：

image-20221002003124276.png

重新聚类后的结果可视化：总共得到34个cluster

sc.pl.umap(adata, color = ['leiden'], frameon = False, legend_loc = "on data")
plt.savefig(dir+"02-intergration_umap_1.png")

## 有多少cluster
adata.obs['leiden']

#Name: leiden, Length: 105264, dtype: category
#Categories (34, object): ['0', '1', '2', '3', ..., '30', '31', '32', '33']

结果图：

image-20221002003157201.png

细胞类型注释

文章中进行了三次注释，第一次注释大类，主要为9个类：

image-20221002202142318.png

这几个类的基因文章中没有提供，就用我们自己收集的基因来进行注释好了。

在视频资源中，视频speaker老师给了一张图：是目前免疫细胞分类很详细的一张图了：

https://learn.cellsignal.com/hubfs/landing-pages/2019/18-IMM-18284/18-IMM_18284-Human%20Markers%20PWHO-digital.pdf

image-20221002215228373.png

基因可视化：

markers = ['EPCAM','KRT8','KRT18','KRT19',  # epithelial cells
          'CD68', 'CTSS', 'FCN1','CD163',   # myeloid cells
          'ACTA2', 'DCN', 'ACTB' , # fibroblasts
          'PECAM1','CD34','VWF',   # endothelial cells
          'PTPRC','CD3D','CD3E','CD3G','CD8A', 'CD4', # T and natural killer (NK) lymphocytes
          'CD79A', 'MS4A1','CD19',  # B lymphocytes and plasma cells
          'SNAP25', 'SYT1',         # neuronal cells
          'CPA3',                   # mast cells
          'CST3', 'LAMP3', 'HLA-DQB2', 'HLA-DPB1', 'BIRC3', # antigen-presenting cells (APCs; primarily dendritic cells) 
          'MKI67','TOP2A', # Cycling
           'HBB','HBD'     # Erythroid
          ]

# 看某一个基因的表达情况
markers[markers.names=="HBB"]
markers[markers.group=="30"]

#, layer = 'scvi_normalized'
# , vmax = 5
sc.pl.umap(adata, color = ["HBB"], frameon = False, layer = 'scvi_normalized', vmax =4)
plt.savefig(dir+"03-intergration_umap_Erythroid_1.png")

一marker为基础，绘制如下类似图，vmax参数可以进行调整让结果看起来更明显，注视不出来的可以看看每个cluster高表达的基因，查一查功能就立马可以推断出来了：

image-20221002003232078.png

结合marker表达以及cluster特异性高表达基因：

image-20221003010745511.png

细胞注释如下：

for x in range(0,35):
    print(f'"{x}":"", ')

cell_type = {"0":"T Cell",
"1":"Epithelial Cell",
"2":"Myeloid Cell",
"3":"Fibroblast",
"4":"Myeloid Cell",
"5":"T Cell",
"6":"Myeloid Cell",
"7":"Epithelial Cell",
"8":"Endothelial",
"9":"Fibroblast",
"10":"Myeloid Cell",
"11":"pDC",
"12":"Epithelial Cell",
"13":"Fibroblast",
"14":"Fibroblast",
"15":"Epithelial Cell",
"16":"Epithelial Cell",
"17":"Myeloid Cell",
"18":"Epithelial Cell",
"19":"Epithelial Cell",
"20":"Neuronal Cell",
"21":"Epithelial Cell",
"22":"B Cell",
"23":"Mast Cell",
"24":"T Cell",
"25":"pDC",
"26":"Myeloid Cell",
"27":"Endothelial",
"28":"Fibroblast",
"29":"Myeloid Cell",
"30":"Erythroid",
"31":"B Cell",
"32":"Neuronal Cell",
"33":"Myeloid Cell"
}

adata.obs['cell type'] = adata.obs.leiden.map(cell_type)
sc.pl.umap(adata, color = ['cell type'], frameon = False, legend_loc = "on data")
plt.tight_layout()
plt.savefig(dir+"04-intergration_umap_anno.png", dpi=300)

adata
adata.uns['scvi_markers'] = markers_scvi
adata.uns['markers'] = markers

# 保存
adata.write_h5ad(dir + 'integrated_anno.h5ad')
model.save(dir + 'model.model')

细胞注释结果如下：

image-20221003011549203.png

这里注释出来了一串文献中没有的红细胞。

下次进行详细注释~~~