Manifold alignment for heterogeneous single-cell multi-omics data integration using Pamona.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell multi-omics sequencing provides a detailed molecular understanding of cells.
• Integrating diverse single-cell multi-omics datasets remains a significant computational challenge.
• Current manifold alignment methods often require datasets to share the same underlying cellular structure.

Purpose of the Study:

• To develop a novel framework, Pamona, for integrating heterogeneous single-cell multi-omics datasets.
• To delineate and represent shared and dataset-specific cellular structures across different data modalities.
• To align cellular modalities within a common low-dimensional space while preserving unique and shared structures.

Main Methods:

• Pamona utilizes a partial Gromov-Wasserstein distance-based manifold alignment approach.
• The framework formulates integration as a partial manifold alignment problem solved via optimal transport.
• It identifies shared and dataset-specific cells using probabilistic couplings and incorporates prior information for enhanced alignment.

Main Results:

• Pamona successfully identifies shared and dataset-specific cells across heterogeneous single-cell multi-omics data.
• The framework accurately recovers and aligns cellular structures from different modalities into a unified low-dimensional space.
• Performance evaluations on benchmark datasets demonstrate Pamona's superiority over existing methods.

Conclusions:

• Pamona offers a robust solution for integrating heterogeneous single-cell multi-omics data.
• The framework effectively captures both shared and modality-specific cellular information.
• Pamona advances the field of single-cell data integration and analysis.