批次校正

38. 批次校正#

38.1. 动机#

正如我们前面对 ADT 数据所做的可视化所示，不同供体之间的批次效应非常明显（见降维）。因此，需要进行批次校正来减轻这种影响。

这里我们使用 Harmony。目前还没有针对 ADT 数据对不同批次校正方法做过基准测试。因此，我们选用 Harmony——它在 scRNA-seq 数据上经过基准测试，效果良好。

最近，有两种针对 ADT 数据的批次校正方法发表在知名期刊上和/或由知名作者提出：ADTnorm [Zheng et al., 2025] 和 CytoVI [Ingelfinger et al., 2025]。这两种方法可能适用于 ADT 数据。不过，如前所述，这里我们坚持使用 Harmony，因为它更成熟、且经过了独立的基准测试（尽管是针对转录组数据，而非 ADT 数据）。

38.2. 环境设置#

import warnings

import muon as mu
import scanpy as sc

warnings.filterwarnings("ignore")
mu.set_options(pull_on_update=False)
sc.settings.verbosity = 0
sc.set_figure_params(
    dpi=80,
    facecolor="white",
    frameon=False,
)

import lamindb as ln

ln.track()

→ found notebook batch_correction.ipynb, making new version
→ created Transform('4LJehi0GPRuj0003', key='batch_correction.ipynb'), started new Run('9GLRcMWow8KIOWYb') at 2026-04-10 17:31:12 UTC
→ notebook imports: lamindb-core==2.3.1 muon==0.1.7 scanpy==1.12
• recommendation: to identify the notebook across renames, pass the uid: ln.track("4LJehi0GPRuj")

38.3. 加载数据#

我们载入上一章降维末尾保存的 MuData 对象：

af = ln.Artifact.connect("theislab/sc-best-practices").get(
    key="surface-protein/cite_dimensionality_reduction.h5mu", is_latest=True
)
mdata = af.load()
mdata

MuData object with n_obs × n_vars = 117951 × 36737
  var:	'gene_ids', 'feature_types'
  2 modalities
    rna:	117951 x 36601
      obs:	'donor', 'batch'
      var:	'gene_ids', 'feature_types'
    prot:	117951 x 136
      obs:	'donor', 'batch', 'n_genes_by_counts', 'log1p_n_genes_by_counts', 'total_counts', 'log1p_total_counts', 'n_counts', 'outliers', 'doublets_markers'
      var:	'gene_ids', 'feature_types', 'n_cells_by_counts', 'mean_counts', 'log1p_mean_counts', 'pct_dropout_by_counts', 'total_counts', 'log1p_total_counts'
      uns:	'batch_colors', 'donor_colors', 'doublets_markers_colors', 'neighbors', 'pca', 'umap'
      obsm:	'X_pca', 'X_umap'
      varm:	'PCs'
      obsp:	'connectivities', 'distances'

38.4. Harmony#

目前尚不清楚哪种批次效应校正方法对 ADT 数据效果最好。对于一般用途，我们推荐 Harmony [Korsunsky et al., 2019] 来对数据进行批次校正，因为它在 scRNA-seq 数据上表现稳健。

sc.external.pp.harmony_integrate(adata=mdata["prot"], key="donor", random_state=0)

2026-04-10 19:31:19,113 - harmonypy - INFO - Computing initial centroids with sklearn.KMeans...
2026-04-10 19:31:26,609 - harmonypy - INFO - sklearn.KMeans initialization complete.
2026-04-10 19:31:26,967 - harmonypy - INFO - Iteration 1 of 10
2026-04-10 19:31:55,536 - harmonypy - INFO - Iteration 2 of 10
2026-04-10 19:32:26,257 - harmonypy - INFO - Iteration 3 of 10
2026-04-10 19:32:56,240 - harmonypy - INFO - Iteration 4 of 10
2026-04-10 19:33:26,641 - harmonypy - INFO - Iteration 5 of 10
2026-04-10 19:33:56,255 - harmonypy - INFO - Converged after 5 iterations

现在，我们基于经 Harmony 校正的 PCA 计算一个邻域图，并计算一个 UMAP 嵌入，以可视化该研究的各个变量。

sc.pp.neighbors(mdata["prot"], n_pcs=20, use_rep="X_pca_harmony", random_state=0)
sc.tl.umap(mdata["prot"], random_state=0)

sc.pl.umap(mdata["prot"], color=["donor", "batch"])

../_images/9fd7deddf5c938f83a3f459a1496e8757c93ca7b0c906bec0266f95810d335e9.png

可以看到，与之前相比，不同供体的细胞在嵌入中的混合程度高了很多（见降维章节）。

sc.pl.umap(mdata["prot"], color=["CD4-1", "CD8", "CD3"])
sc.pl.umap(mdata["prot"], color=["CD14-1", "CD16"])

../_images/df62334659752670a6b91a4f75dccd010438817855a8f2141c518226738f2409.png

../_images/4c8920e87659102f88364566f77caa1ecc950ea4df68cd71ab6c3291221f2960.png

我们检查几个标记基因的表达，以确认不同的细胞类型之间仍然彼此分开。可以看到，T 细胞仍然形成一个独立的群体，并进一步分为 CD4 和 CD8 T 细胞。此外，与降维之前不同的是，现在 CD4 T 细胞形成了一个离散的聚类，其中各供体相互混合。因此，批次校正是成功的。

af_batch_correction = ln.Artifact.from_mudata(
    mdata,
    key="surface-protein/cite_batch_correction.h5mu",
    description="CITE-seq data after batch correction",
)
af_batch_correction.save()

ln.finish()

38.5. 参考文献#

[spILE+25]

Florian Ingelfinger, Nathan Levy, Can Ergen, Artemy Bakulin, Alexander Becker, Pierre Boyeau, Martin Kim, Diana Ditz, Jan Dirks, Jonas Maaskola, and others. Cytovi: deep generative modeling of antibody-based single cell technologies. bioRxiv, pages 2025–09, 2025.

[spKMF+19]

Ilya Korsunsky, Nghia Millard, Jean Fan, Kamil Slowikowski, Fan Zhang, Kevin Wei, Yuriy Baglaenko, Michael Brenner, Po-ru Loh, and Soumya Raychaudhuri. Fast, sensitive and accurate integration of single-cell data with harmony. Nature Methods, 16(12):1289–1296, Dec 2019. URL: https://doi.org/10.1038/s41592-019-0619-0, doi:10.1038/s41592-019-0619-0.

[spZCK+25]

Ye Zheng, Daniel P Caron, Ju Yeong Kim, Seong-Hwan Jun, Yuan Tian, Florian Mair, Kenneth D Stuart, Peter A Sims, and Raphael Gottardo. Adtnorm: robust integration of single-cell protein measurement across cite-seq datasets. Nature communications, 16(1):5852, 2025.

38.6. 贡献者#

我们衷心感谢以下人员的贡献：

38.6.1. 作者#

Javier Marchena-Hurtado
Daniel Strobl
Ciro Ramírez-Suástegui

38.6.2. 审阅者#

Lukas Heumos
Anna Schaar