使用例

#### API ### # sc.pp.normalize_per_cell # scv.pp.normalize_per_cell .obs.n_counts .var.gene_count_corr # sc.pp.log1p .uns.log1p # scv.pp.log1p # sc.pp.highly_variable_genes .var.highly_variable .var.means .var.dispersions .var.dispersions_norm .uns.hvg # sc.pp.scale .var.mean .var.std # scvi 明示的に .obs.solo .obsm.X_scVI .layers.scVI_normalized # sc.pp.neighbors .uns.neighbors .obsp.distances .obsp.connectivities # sc.tl.umap .uns.umap .obsm.X_umap # sc.tl.leiden .uns.leiden .obs.leiden_scVI # 明示的に

import

import os
import numpy as np
import pandas as pd
import scanpy as sc

データ入力

  #adata = sc.read("adata.h5ad")
adata = sc.read_10x_mtx(10X_dir, var_names="gene_symbols")
adata.var_names_make_unique()
adata = adata[:, adata.X.sum(axis=0).A1 != 0]
  #AnnData Class
  # .X    data matrix
  # .var  annotation of variables
  # .obs  annotation of observations
  # .uns  unstruct annotation

metrics / subset

Mt = adata.var_names.str.startswith('mt-') # NEED_TO_CHAGE
adata.obs['pMt' ] = np.sum(adata[:, Mt].X, axis = 1).A1 / np.sum(adata.X, axis = 1).A1 * 100
adata.obs['nUMIs' ] = adata.X.sum(axis = 1).A1
adata.obs['nGenes'] = (adata.X > 0.0).astype(int).sum(axis = 1).A1

sc.pl.violin(adata, 
  ['nGenes', 'nUMIs', 'pMt'], 
	multi_panel = True
)

adata = adata[adata.obs['pMt'] < 20, :]

sc.pl.violin(adata, 
  ['nGenes', 'nUMIs', 'pMt'], 
	multi_panel = True
)

NORMALIZE / HVGs / SCALE

sc.pp.normalize_total(adata, target_sum = 1e4)
sc.pp.log1p(adata)
sc.pp.highly_variable_genes(adata, n_top_genes = 2000)
adata.raw = adata
sc.pp.scale(adata)

PCA

sc.tl.pca(adata, svd_solver = "arpack")
sc.pl.pca_variance_ratio(adata, log = True)

graph/dimentional reduction

sc.pp.neighbors(adata, n_neighbors=25, n_pcs=20, use_rep="X_scVI")
  # n_neighbors: int (default: 15)
  # : The size of local neighborhood used for manifold approximation.
  # : Larger values result in more global view of the manifold, while
  # : smaller values result in more local data being preserved.
sc.tl.umap(adata, min_dist=0.4, spread=0.6)
  # min_dist : float (default: 0.5)
  # : Smaller value lead to more clustered/clumped embedding
  # spread : float (default: 1.0)
  # : Larger values result in more clustered/clumped embedding
sc.tl.leiden(adata, resolution=0.4, key_added="leiden_scVI")
  # resolution : float (default: 1)
  # : Larger values result in more clusters.

plot

sc.pl.umap(
  adata, 
  color = 'clusters', 
  frameon = False, 
  size = 20, 
  use_raw = False)
  # raw属性が格納されていると、そのデータが優先される

sc.pl.violin(
  adata, 
  ['gene1', 'gene2'], 
  groupby='clusters' ,
  size = 1.5, 
  xlabel = None,
  ylabel = None 
)

All_markers

sc.tl.rank_genes_groups(adata, 'leiden', method='wilcoxon')
df = pd.DataFrame(adata.uns['rank_genes_groups']['names'])

Rename

D = {
  '0': 'cell_1',
  '1': 'cell_2',
  '2': 'cell_2',
  '3': 'cell_1'
}

adata.obs['NEW'] = (
  adata.obs['leiden_scVI']
  .map(D)
  .astype('category')
)

save

adata.write("adata.h5ad")
df.to_csv("df.csv")
sc.pl.umap(adata, save="_sufix.png") # saved in "figures" dir

参考文献