PalmaClust: A graph-fusion framework leveraging the Palma ratio for robust ultra-rare cell type detection in...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is vital for tissue atlases, developmental studies, and disease microenvironment characterization.
• Standard scRNA-seq clustering pipelines often fail to detect rare cell populations (e.g., progenitors, resistant tumor cells, specific lymphocytes) due to their low frequencies (<1%).
• Existing methods require manual curation, rely on known markers, or suffer from high false positive rates, necessitating a more sensitive and statistically grounded approach.

Purpose of the Study:

• To develop a scalable and statistically robust method for sensitive detection of rare cell populations in scRNA-seq data.
• To provide calibrated confidence and interpretable molecular signatures for identified rare cell populations.
• To overcome the limitations of existing clustering pipelines in identifying low-frequency cell types.

Main Methods:

• Introduction of PalmaClust, a graph-fusion clustering framework.
• Repurposing the Palma ratio, a tail-sensitive inequality metric, to identify marker genes associated with extreme sparsity.
• Construction and fusion of multiple K-Nearest Neighbor (KNN) graphs using complementary gene-selection statistics (Palma ratio, Gini index, Fano factor).
• Implementation of a local refinement strategy to re-prioritize Palma-ranked genes within parent clusters.

Main Results:

• PalmaClust consistently outperforms state-of-the-art methods across diverse scRNA-seq datasets.
• Significant improvement in rare-class F1 scores by at least 20% (absolute) compared to existing baselines.
• Maintenance of high global clustering stability alongside enhanced rare population detection.
• Demonstration that the Palma ratio-derived graph layer is crucial for capturing ultra-rare cell signatures missed by other methods.

Conclusions:

• PalmaClust offers a statistically grounded and scalable solution for sensitive rare cell population detection in scRNA-seq data.
• The framework provides calibrated confidence and interpretable molecular signatures, advancing biological and clinical insights.
• The novel use of the Palma ratio in graph construction is key to identifying previously undetectable cell populations.