单细胞ATAC实战03: 使用SnapATAC进行降维、批次校正、聚类

生信技能树jimmy

发布于 2023-09-26 20:20:52

3470

发布于 2023-09-26 20:20:52

from pathlib import Path
import warnings

import numpy as np
import pandas as pd
import scanpy as sc
import snapatac2 as snap

import bioquest as bq
import sckit as sk

基因组注释文件

gff_file="~/DataHub/Genomics/GENCODE/hg38.v43.chr_patch_hapl_scaff.basic.annotation.gff3.gz"

读取数据

h5ad_files = []
for i in Path("data").glob("**/fragments.tsv.gz"):
    _s = str(i).split('/')[1]
    print(_s)
    _a = snap.pp.import_data(fragment_file=i, gff_file=gff_file,
                             chrom_size=sk.utils.Chrom_size.hg38, min_tsse=7, min_num_fragments=1000)
# create a cell by bin matrix containing insertion 
# counts across genome-wide 500-bp bins.
    snap.pp.add_tile_matrix(_a)
# feature selection
    snap.pp.select_features(_a)

# Doublet removal
    snap.pp.scrublet(_a)
    snap.pp.call_doublets(_a)

    h5ad_filename = f"{_s}.h5ad"
    _a.write(h5ad_filename)
    h5ad_files.append((_s, h5ad_filename))
# pbmc_5k
# 2023-03-27 20:01:46 - INFO - Simulating doublets...
# 2023-03-27 20:01:47 - INFO - Spectral embedding ...
# 2023-03-27 20:01:54 - INFO - Calculating doublet scores...
# pbmc_10k
# 2023-03-27 20:10:42 - INFO - Simulating doublets...
# 2023-03-27 20:10:43 - INFO - Spectral embedding ...
# 2023-03-27 20:11:21 - INFO - Calculating doublet scores...

合并两个对象

AnnDataSet不同于anndata的是，AnnDataSet关联了两个本地文件pbmc_5k.h5ad和pbmc_10k.h5ad，因此在不用这个对象的时候需要手动关闭adata.close()，类似open函数。

dataset = snap.AnnDataSet(adatas=h5ad_files, filename="pbmc.h5ads")
dataset.obs_names = dataset.obs['sample'].to_numpy() + ":" + np.array(dataset.obs_names)
dataset
# AnnDataSet object with n_obs x n_vars = 15957 x 6176550 backed at 'pbmc.h5ads'
# contains 2 AnnData objects with keys: 'pbmc_5k', 'pbmc_10k'
#     obs: 'sample'
#     var: 'selected'
#     uns: 'AnnDataSet'

snap.pp.select_features(adata)

dimension reduction

降维

randomly selects 10,000 cells(sample_size) as the landmarks to get an initial embedding, and then uses the Nystrom method to compute the embedding for the rest of cells.

snap.tl.spectral(adata, sample_size = 10000)
snap.pl.spectral_eigenvalues(adata, interactive=False)

snap.tl.umap(adata, use_dims = 12)
snap.pl.umap(adata, color="sample", interactive=False, width = 800)

Batch correction

使用harmony去除批次效应

snap.pp.harmony(adata, "sample", use_dims = 12, max_iter_harmony = 20)
snap.tl.umap(adata, use_rep="X_spectral_harmony")
snap.pl.umap(adata, color="sample", interactive=False, width=800)

snap.pp.harmony(adata, "sample", use_dims = 12, max_iter_harmony = 20)
snap.tl.umap(adata, use_rep="X_spectral_harmony")
snap.pl.umap(adata, color="sample", interactive=False, width=800)

Clustering

使用leiden对细胞聚类

snap.pp.knn(adata, use_rep="X_spectral_harmony", use_dims = 12)
snap.tl.leiden(adata, resolution = 1.0)
snap.pl.umap(adata, color="leiden", interactive=False, width=600)