作者：童蒙
编辑：angelica

scanpy代码解读来啦~

单细胞分析第一步是对数据进行标准化，标准化的方法有很多，下面给大家解读一下scanpy的一个：函数为：scanpy.pp.normalize_total

作用

对每个细胞的count进行标准化，使得每个细胞标准化后有相同的count

如果使用target_sum=1e6，那么相当于CPM标准化。
设置exclude_highly_expressed=True 后，非常高的表达基因会被排除到计算size factor中，这个会影响其他的基因。需要同max_fraction进行连用，只要有一个细胞里面超过了，就会被判别为高表达基因。默认为0.05。

常见参数

adata：内置的AnnDta数据
target_sum: 如果设置为none，那么会用所有样品的median值来替代；
exclude_highly_expressed ：是否去除高表达基因
max_fraction：高表达基因的阈值
key_added ：是否再obs里面加一个属性
layer：针对哪一个layer

案例

from anndata import AnnData
import scanpy as sc
sc.settings.verbosity = 2
np.set_printoptions(precision=2)
adata = AnnData(np.array([
   [3, 3, 3, 6, 6],
   [1, 1, 1, 2, 2],
   [1, 22, 1, 2, 2],
]))
adata.X
array([[ 3.,  3.,  3.,  6.,  6.],
       [ 1.,  1.,  1.,  2.,  2.],
       [ 1., 22.,  1.,  2.,  2.]], dtype=float32)
X_norm = sc.pp.normalize_total(adata, target_sum=1, inplace=False)['X']
X_norm
array([[0.14, 0.14, 0.14, 0.29, 0.29],
       [0.14, 0.14, 0.14, 0.29, 0.29],
       [0.04, 0.79, 0.04, 0.07, 0.07]], dtype=float32)
X_norm = sc.pp.normalize_total(
    adata, target_sum=1, exclude_highly_expressed=True,
    max_fraction=0.2, inplace=False
)['X']
The following highly-expressed genes are not considered during normalization factor computation:
['1', '3', '4']
X_norm
array([[ 0.5,  0.5,  0.5,  1. ,  1. ],
       [ 0.5,  0.5,  0.5,  1. ,  1. ],
       [ 0.5, 11. ,  0.5,  1. ,  1. ]], dtype=float32)

代码解读

对特定的代码进行重点介绍一下，有以下三个：

对于高表达基因的确定

if exclude_highly_expressed:
    counts_per_cell = adata.X.sum(1)  # original counts per cell
    counts_per_cell = np.ravel(counts_per_cell)

    # at least one cell as more than max_fraction of counts per cell
    gene_subset = (adata.X > counts_per_cell[:, None] * max_fraction).sum(0)
    gene_subset = np.ravel(gene_subset) == 0

    msg += (
        ' The following highly-expressed genes are not considered during '
        f'normalization factor computation:\n{adata.var_names[~gene_subset].tolist()}'
    )

确定size factor

counts_per_cell = X.sum(1)
counts_per_cell = np.ravel(counts_per_cell).copy()
adata.X = _normalize_data(adata.X, counts_per_cell, target_sum)

标准化

def _normalize_data(X, counts, after=None, copy=False):
    X = X.copy() if copy else X
    if issubclass(X.dtype.type, (int, np.integer)):
        X = X.astype(np.float32)  # TODO: Check if float64 should be used
    counts = np.asarray(counts)  # dask doesn't do medians
    after = np.median(counts[counts > 0], axis=0) if after is None else after
    counts += counts == 0
    counts = counts / after
    if issparse(X):
        sparsefuncs.inplace_row_scale(X, 1 / counts)
    else:
        np.divide(X, counts[:, None], out=X)
    return X

相信大家看了代码，能够理解内部的运行方式，请继续关注我们吧。

参考资料

https://scanpy.readthedocs.io/en/stable/generated/scanpy.pp.normalize_total.html

代码解读- scanpy.pp.normalize_total