scanorama

Scanorama: Single-Cell Data Integration

Overview

Scanorama is a Python package for integrating multiple single-cell datasets, particularly useful for batch correction and combining data from different sources. It uses an iterative strategy to align shared cell types across datasets while preserving dataset-specific populations.

When to Use This Skill

This skill should be used when:

• Integrating multiple single-cell datasets in Python
• Performing batch correction on scRNA-seq data
• Combining datasets from different batches or technologies
• Integrating datasets with partial overlap in cell types
• Working with large-scale datasets (scales well)
• Removing technical noise while preserving biological variation

Quick Start

Installation

# Install via pip
pip install scanorama

# Or from GitHub
pip install git+https://github.com/brianhie/scanorama.git

Basic Integration

import scanpy as sc
import scanorama

# Load multiple datasets
adata1 = sc.read_h5ad('batch1.h5ad')
adata2 = sc.read_h5ad('batch2.h5ad')
adata3 = sc.read_h5ad('batch3.h5ad')

# Put in list
adatas = [adata1, adata2, adata3]

# Integration
corrected = scanorama.correct_scanpy(adatas, return_list=True)

# Combine corrected datasets
adata_combined = sc.concat(corrected)

# Update obs with batch info
for i, ad in enumerate(corrected):
    ad.obs['batch'] = f'batch{i}'

# Continue with standard analysis
sc.pp.neighbors(adata_combined)
sc.tl.umap(adata_combined)
sc.pl.umap(adata_combined, color='batch')

Integration Workflow

Full Example

import scanpy as sc
import scanorama
import numpy as np

# Load datasets
datasets = []
labels = []

for batch in ['batch1', 'batch2', 'batch3']:
    adata = sc.read_h5ad(f'{batch}.h5ad')
    datasets.append(adata)
    labels.extend([batch] * adata.n_obs)

# Preprocess each dataset
for adata in datasets:
    sc.pp.normalize_total(adata, target_sum=1e4)
    sc.pp.log1p(adata)
    sc.pp.highly_variable_genes(adata, n_top_genes=2000)

# Integrate
corrected, genes = scanorama.correct(datasets, return_dimred=True)

# Create combined AnnData
adata_combined = scanorama.assemble_scanpy(corrected)

# Add batch labels
adata_combined.obs['batch'] = labels

# PCA and UMAP
sc.tl.pca(adata_combined, n_comps=50)
sc.pp.neighbors(adata_combina

, n_neighbors=15, n_pcs=50)
sc.tl.umap(adata_combined)

# Visualization
sc.pl.umap(adata_combined, color='batch')