Bridging unpaired single-cell multimodal data for integrative analyses with SuperMap.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell profiling technologies generate multimodal data, offering deep biological insights.
• Most multimodal single-cell datasets are unpaired, lacking cell-wise correspondence, hindering integration.
• Integrating unpaired multimodal data is crucial for understanding complex biological systems.

Purpose of the Study:

• To introduce SuperMap, a novel statistical learning method for analyzing unpaired multimodal single-cell data.
• To develop a method that directly learns cross-modal mappings from unpaired data.
• To facilitate downstream analysis tasks including cell-type identification and trajectory inference.

Main Methods:

• SuperMap employs statistical learning to directly map cross-modal features from unpaired data.
• The method establishes linkages between distinct modalities without requiring paired measurements.
• Benchmarking involved comprehensive evaluations on simulated and real-world datasets.

Main Results:

• SuperMap demonstrated superior performance in integrating unpaired multimodal single-cell data.
• The method significantly enhanced cell-type identification accuracy.
• SuperMap enabled improved diagonal integration, regulatory analysis, and revealed epigenomic priming events for trajectory inference.

Conclusions:

• SuperMap effectively addresses the challenge of integrating unpaired multimodal single-cell data.
• The method provides a powerful tool for bridging different cellular modalities.
• SuperMap facilitates deeper biological interpretation and discovery from complex single-cell datasets.