在scanpy中直接加載稀疏矩陣相比加載密集矩陣可以幫助節(jié)省數(shù)據(jù)加載的內(nèi)存占用

from scipy.io import mmread
import scipy.sparse as sp
import anndata as ad
import joblib
#加載數(shù)據(jù)
X = mmread( 'matrix.mtx.gz')
X = X.T
X = sp.csr_matrix(X)  #將讀取的coo_matrix 轉(zhuǎn)換為scanpy對(duì)象的crs_matrix(按行壓縮的sparse matrix)
X.data = np.log2( X.data + 1 )  #對(duì)crs_matrix矩陣做標(biāo)準(zhǔn)化,非必須
obs = pd.read_csv('metaData.tsv',sep = '\t') #obs信息添加前面分析的細(xì)胞meta.data數(shù)據(jù)
var = pd.read_csv( 'features.tsv', index_col=0,sep = '\t',header=None) #基因信息
var.index.name = "gene_id"
ad_sc = ad.AnnData(X = X, obs = obs, var = var)
joblib.dump(ad_sc ,"ad_sc.pkl") #打包出工程對(duì)象(讀取: joblib.load("ad_sc.pkl"))