UnionCom

UnionCom aligns single-cell multi-omics datasets by unsupervised topological alignment to integrate unpaired cells and unmatched features for comparative analysis of cellular heterogeneity and molecular mechanisms.

Key Features:

• Unsupervised Topological Alignment: Aligns datasets without requiring any prior correspondence at the cellular or feature level.
• Distance Matrix Embedding and Optimization: Embeds the intrinsic low-dimensional structure of each dataset into distance matrices of cell proximities and aligns datasets by matching these matrices via matrix optimization.
• Common Feature Space Projection: Projects distinct and unmatched features from different single-cell datasets into a shared embedding space to enable feature comparability among aligned cells.
• Global Scaling Parameter Adjustment: Introduces a global scaling parameter to adjust distance matrices for matching complex nonlinear geometrical structures across datasets.
• Robustness and Flexibility: Demonstrates robustness to parameter choices and subsampling of features and can accommodate samples with dataset-specific cell types.

Scientific Applications:

• Single-cell multi-omics integration: Integrates unpaired single-cell multi-omics datasets for joint analysis across modalities.
• Cross-dataset cell alignment and comparison: Enables alignment and comparison of cells across datasets without requiring one-to-one correspondence.
• Characterization of cellular heterogeneity: Facilitates investigation of cellular heterogeneity and underlying molecular mechanisms.
• Evaluation on simulated and real datasets: Applicable to both simulated and real-world biological datasets for analysis and method assessment.

Methodology:

UnionCom uses an unsupervised workflow that embeds each dataset's intrinsic low-dimensional structure into distance matrices of cell proximities, matches these distance matrices across datasets via matrix optimization, introduces a global scaling parameter to adjust distances, and projects features into a common embedding space.