Pamona

Pamona performs partial manifold alignment of heterogeneous single-cell multi-omics datasets to identify shared and dataset-specific cellular structures and enable integrative analysis across different experimental conditions and platforms.

Key Features:

• Partial Manifold Alignment: Formulates integration as a partial manifold alignment problem to separate shared and dataset-specific cellular structures.
• Partial Gromov-Wasserstein Optimal Transport: Implements a partial Gromov-Wasserstein optimal transport framework to compute probabilistic couplings between cells across datasets.
• Probabilistic Couplings for Cell Identification: Uses probabilistic couplings to identify cells common across datasets (shared) and cells unique to specific datasets (dataset-specific).
• Alignment in a Common Low-Dimensional Space: Aligns cellular modalities into a unified low-dimensional space while preserving shared and dataset-specific structures.
• Incorporation of Prior Information: Integrates prior information such as cell type annotations or known cell-cell correspondences to refine alignment quality.

Scientific Applications:

• Multi-omics integration across conditions and platforms: Integrates heterogeneous single-cell multi-omics datasets from different experimental conditions or platforms for comparative analyses.
• Study of cellular heterogeneity and disease mechanisms: Identifies shared and dataset-specific structures to explore molecular mechanisms underlying cellular heterogeneity, disease progression, and treatment responses.

Methodology:

Formulates integration as a partial manifold alignment problem, employs a partial Gromov-Wasserstein optimal transport framework to compute probabilistic couplings, aligns cells in a common low-dimensional space, and incorporates prior information such as cell type annotations or known cell-cell correspondences.