Optimal transport improves cell-cell similarity inference in single-cell omics data

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• High-throughput single-cell profiling reveals cellular heterogeneity in development and disease.
• Unsupervised clustering is key for identifying cell types and states.
• Clustering quality depends heavily on the chosen cell-cell similarity metric.

Purpose of the Study:

• To introduce Optimal Transport (OT) as a novel cell-cell similarity metric for single-cell omics data.
• To evaluate the performance of OT against existing metrics.
• To demonstrate the utility of OT for inferring cellular similarity and improving clustering.

Main Methods:

• Applied entropic regularization to Optimal Transport for computational efficiency in high-dimensional single-cell data.
• Benchmarked OT against state-of-the-art metrics using 13 diverse datasets (simulated, scRNA-seq, scATAC-seq, DNA methylation).
• Assessed metric performance in detecting cell group similarity and evaluated clustering quality.

Main Results:

• Optimal Transport (OT) significantly improved cell-cell similarity inference across all tested single-cell omics datasets.
• OT enhanced the quality of unsupervised clustering for simulated, scRNA-seq, scATAC-seq, and DNA methylation data.
• The proposed OT-based approach demonstrated superior performance compared to existing similarity metrics.

Conclusions:

• Optimal Transport provides a robust and effective metric for analyzing single-cell omics data.
• The OT-scOmics framework offers a reproducible method for advancing single-cell data analysis.
• This work facilitates more accurate characterization of cellular heterogeneity in biological and medical research.