mnnCorrect

mnnCorrect is an algorithm designed to correct batch effects in gene expression data, with a focus on single-cell RNA sequencing. It implements the mutual nearest neighbors (MNN) approach, which aligns cells across batches by identifying corresponding cell states and removing technical variation while preserving biological signals.

The method works by comparing batches in a reduced expression space. After normalizing and log-transforming the data (often after selecting a set of highly variable genes), mnnCorrect searches for k-nearest neighbors across batches and identifies mutual pairs—cells that are each other’s mutual nearest neighbors. The differences in expression between these matched pairs are used to estimate batch-specific correction vectors. These vectors are then aggregated and applied to adjust the expression values, producing a corrected dataset suitable for downstream analyses such as clustering, dimensionality reduction, and trajectory inference.

mnnCorrect can be used to merge two batches or multiple batches, either sequentially or in a multi-batch framework. It is implemented in Bioconductor, commonly via the scran package, and is frequently used in pipelines that standardize preprocessing steps like normalization and variable gene selection before integration. The approach is valued for its ability to remove batch effects without explicit modeling of batch covariates and for its robustness to differing cell compositions.

Limitations and considerations include the requirement that batches share overlapping cell states; if biology and batch effects are confounded, correction may be less effective. It can be computationally intensive for very large datasets, though newer implementations (such as fastMNN) address scalability. Users should consider gene selection, normalization methods, and the order of batch merging to achieve optimal alignment.