Batch Effect Correction

An iterative algorithm that performs soft clustering of cells across batches and applies a mixture model-based correction. It does not require a pre-defined anchor set, making it highly scalable to large atlases.

• Mechanism: Iteratively clusters cells, calculates cluster-specific linear correction factors, and repeats until convergence.
• Strengths: Fast, memory-efficient, and integrates seamlessly with the Seurat ecosystem.
• Use Case: Integrating 10+ patient samples for a population-level reference atlas.

A foundational method that identifies pairs of cells from different batches that are mutual nearest neighbors in high-dimensional space. It estimates the batch vector based on the differences between these paired cells.

• Mechanism: Finds robust cell pairs across batches, calculates the delta in expression, and applies a smoothing kernel to correct the entire dataset.
• Strengths: Conceptually simple and effective for simple experimental designs.
• Limitation: Can be computationally intensive for very large datasets and struggles with complex, nested batch structures.

A deep generative model based on a variational autoencoder that learns a non-linear, probabilistic latent representation of gene expression. It explicitly models batch effects as a latent variable.

• Mechanism: Uses a neural network to encode gene expression into a low-dimensional space while conditioning on batch labels, then decodes to reconstruct the original counts.
• Strengths: Captures complex non-linear batch effects, handles zero-inflation naturally, and provides uncertainty estimates.
• Output: A corrected latent space suitable for clustering and trajectory inference.

A canonical correlation analysis (CCA)-based method that identifies anchors—shared cell pairs across datasets—to compute a transformation matrix. It is the default integration workflow in the Seurat R package.

• Mechanism: Projects datasets into a shared low-dimensional space using CCA, finds mutual nearest neighbors as anchors, and scores the consistency of these anchors.
• Strengths: Robust to diverse data types (scRNA-seq, scATAC-seq) and allows for the transfer of labels and metadata from a reference to a query.
• Workflow: FindIntegrationAnchors → IntegrateData.

An adaptation of the classic ComBat method for bulk data, specifically tailored for negative binomial distributed single-cell RNA-seq counts. It adjusts the mean and variance of gene expression across batches.

• Mechanism: Uses an empirical Bayes framework to shrink batch effect estimates, preserving the integer nature of count data.
• Strengths: Fast, interpretable, and outputs a corrected count matrix that can be used directly for downstream differential expression analysis.
• Assumption: Assumes a linear additive batch effect, which may not capture complex non-linear technical variation.

A graph-based method that constructs a k-nearest neighbor graph by connecting each cell to its nearest neighbors in each batch separately, rather than globally. It is a preprocessing step for graph-based clustering.

• Mechanism: Builds a neighborhood graph that is balanced across batches, effectively forcing the clustering algorithm to ignore batch-specific groupings.
• Strengths: Extremely fast and scalable to millions of cells, integrated into the Scanpy ecosystem.
• Critical Note: BBKNN outputs a corrected graph, not a corrected expression matrix. It is not suitable for workflows requiring a corrected feature space, like trajectory inference.