Mutual Nearest Neighbors

Mutual Nearest Neighbors (MNN) is a robust batch-correction algorithm that identifies shared cell states across high-dimensional genomic datasets to eliminate technical noise.

MNN solves the batch effect problem in single-cell RNA sequencing by identifying pairs of cells that are each other's closest neighbors across different experiments. Unlike global alignment methods that risk over-correcting biological signal, MNN uses local anchors to compute a correction vector (the difference between paired expression profiles). This approach preserves unique cell types present in only one sample while merging shared populations. It is the gold standard for integrating diverse datasets, such as combining 10x Genomics and Drop-seq runs, ensuring that downstream clustering reflects biology rather than laboratory bias.

https://www.nature.com/articles/nbt.4091