FR-Match

FR-Match applies the Friedman-Rafsky non-parametric test to perform cluster-to-cluster matching of cell types across single-cell RNA sequencing (scRNAseq) datasets, enabling integration of experiments and identification of shared and novel cell phenotypes.

Key Features:

• Cluster-to-Cluster Matching: Matches cell type clusters rather than individual cells to assess phenotypic heterogeneity across datasets.
• Supervised Feature Selection: Employs supervised feature selection for dimensionality reduction to prioritize relevant gene expression features during matching.
• Shared Information Integration: Incorporates shared information among cells to determine whether two clusters share the same underlying multivariate gene expression distribution.
• Benchmarking and Validation: Benchmarked against cell-to-cell and cell-to-cluster matching methods using simulated and real scRNAseq data, demonstrating fewer erroneous matches of distinct cell subtypes.
• Identification of Novel Cell Phenotypes: Identifies novel cell phenotypes in new datasets by flagging clusters without matches in reference data.
• Robustness to True Negatives: In silico validation shows robustness to increasing numbers of true negatives (non-represented cell types).

Scientific Applications:

• Human cortical layer 1 analysis: Applied to human brain scRNAseq sampled from cortical layer 1, recapitulating laminar characteristics of matched cell type clusters.
• Middle temporal gyrus mapping: Applied to full-thickness middle temporal gyrus scRNAseq to map cell types across overlapping human brain regions and reflect neuroanatomical distributions.

Methodology:

Applies supervised feature selection for dimensionality reduction and uses the Friedman-Rafsky non-parametric test on shared cell information to assess whether clusters share the same multivariate gene expression distribution.