scLKME: A Landmark-based Approach for Generating Multi-cellular Sample Embeddings from Single-cell Data.

Area of Science:

• Computational Biology
• Immunology
• Genomics

Background:

• Single-cell technologies provide high-dimensional data for cell-type variation analysis.
• Heterogeneous cellular composition across samples complicates per-sample representation.
• Developing methods for meaningful per-sample embeddings is crucial.

Approach:

• scLKME is a novel approach using landmark-based kernel mean embedding.
• It converts multi-sample single-cell data into compact per-sample embeddings.
• Samples are treated as distributions over cells, mapped into a reproducing kernel Hilbert space.

Key Points:

• scLKME outperforms existing methods in robustness, efficiency, and accuracy.
• It generates practical and informative sample embeddings.
• The method accurately identifies patient-specific variations in complex datasets.

Conclusions:

• scLKME offers a powerful tool for analyzing multi-sample single-cell data.
• Its application in immunology, particularly in preterm birth studies, demonstrates broad utility.
• The open-sourced package facilitates accessibility and further research.