ClusterFoldSimilarity

ClusterFoldSimilarity calculates a similarity coefficient using the fold changes of shared features (e.g. genes) among clusters of different samples, batches, or datasets to enable comparison and matching of cell clusters across single-cell experiments.

Key Features:

• Cross-sample cluster comparison: Calculates similarity between clusters from different samples, batches, or datasets using shared features.
• Fold change-based similarity metric: Uses fold changes of features rather than raw expression values to compare clusters.
• Dot-product similarity coefficient: Computes similarity via the dot-product (Hadamard product) of pairwise fold change combinations between source and target clusters.

Scientific Applications:

• Single-cell multi-sample integration: Matches equivalent cell clusters across different single-cell samples or datasets.
• Batch effect assessment: Identifies corresponding clusters across batches to evaluate technical variation versus biological differences.
• Cross-dataset cell type annotation: Facilitates annotation transfer by quantifying cluster similarity between reference and query datasets.

Methodology:

The similarity coefficient is calculated using the dot-product (Hadamard product) of every pairwise combination of fold changes between a source cluster i of sample/dataset n and all target clusters j in sample/dataset m.