anndata

예상 결과:

• Create the expression matrix (X) with shape (observations, variables)
• Create observation metadata DataFrame with cell-level annotations (cell_type, sample, condition)
• Create variable metadata DataFrame with gene-level information (gene_name, chromosome, highly_variable status)
• Pass all components to ad.AnnData constructor with X, obs, and var parameters
• Access n_obs and n_vars properties to verify dimensions match your metadata

예상 결과:

• Use backed mode (backed='r') to access data without loading entire matrix into memory
• Convert sparse data to sparse matrix format (csr_matrix) for 10-100x memory reduction
• Convert string columns to categorical type using astype('category') or strings_to_categoricals()
• Store raw data before filtering using adata.raw = adata.copy()
• Use compression='gzip' when writing h5ad files for smaller storage