SCOT

SCOT performs unsupervised alignment of single-cell multi-omics datasets using Gromov-Wasserstein optimal transport to map heterogeneous measurements into a shared space for integrative analysis.

Key Features:

• Unsupervised learning: Operates without requiring cell-to-cell correspondence or labeled matches across modalities.
• Gromov-Wasserstein optimal transport: Aligns datasets by comparing and matching intra-dataset similarity structures across modalities using Gromov-Wasserstein distance.
• Reduced hyperparameter tuning: Uses Gromov-Wasserstein distance to guide parameter selection and reduce the need for extensive hyperparameter tuning.
• Performance and efficiency: Demonstrates comparable alignment performance to leading methods on simulated and real-world datasets while improving computational speed and scalability.

Scientific Applications:

• Developmental biology: Integrates single-cell multi-omics datasets to reconstruct cell states and transitions during development.
• Oncology: Maps tumor cellular heterogeneity and cell–cell interactions by aligning diverse single-cell measurements.
• Systems biology: Builds comprehensive cellular maps from multi-omics single-cell data for systems-level analyses of cellular states and interactions.

Methodology:

Implemented in Python; performs unsupervised alignment of heterogeneous single-cell multi-omics datasets into a shared space using Gromov-Wasserstein optimal transport, considering intra-dataset similarities and inter-dataset relationships and using the Gromov-Wasserstein distance to guide parameter selection.