A novel coarsened graph learning method for scalable single-cell data analysis.

Area of Science:

• Computational Biology
• Bioinformatics
• Data Science

Background:

• Single-cell technologies generate vast, complex datasets.
• Graph-based analyses are effective but computationally challenging for large single-cell data.
• Existing coarsening methods (Cytocoarsening, HEM, LVE) are slow and lack adaptability.

Purpose of the Study:

• To develop a scalable and efficient method for single-cell data analysis.
• To address the computational challenges of managing large-scale graph representations.
• To improve processing speed and preserve essential data features in single-cell datasets.

Main Methods:

• Feature-Aware Graph Coarsening via Hashing (FACH) integrates locality-sensitive hashing.
• FACH directly extracts informative, low-dimensional cell representations from raw data.
• The method is applied to single-cell RNA sequencing and mass cytometry data.

Main Results:

• FACH significantly improves processing speed compared to existing methods (at least 50% reduction in computational time).
• The method preserves critical biological features, including transcriptional signatures and network topology.
• Achieved superior classification accuracy (e.g., 88.1% on the Baron Human dataset).

Conclusions:

• FACH provides an efficient and precise approach for large-scale single-cell data analysis.
• The method enables scalable graph neural network training on coarsened single-cell data.
• FACH effectively retains crucial biological information for accurate downstream tasks.