Cross-modal contrastive learning decodes developmental regulatory features through chromatin potential analysis.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Multimodal single-cell data, measuring chromatin accessibility and transcription simultaneously, offers a unified approach for regulatory analysis.
• Integrating diverse omics data is crucial for a comprehensive understanding of cellular mechanisms.

Purpose of the Study:

• To introduce Attune, a cross-modal contrastive learning framework for aligning gene expression and chromatin accessibility data.
• To demonstrate Attune's effectiveness in omics integration, gene expression prediction, and regulatory inference.

Main Methods:

• Developed Attune, a cross-modal contrastive learning framework.
• Employed transformer-based cross-modal attention for inferring regulatory interactions.
• Benchmarked Attune's performance against existing methods for omics integration and prediction.

Main Results:

• Attune demonstrated superior performance in integrating multimodal omics data and predicting gene expression.
• The framework successfully inferred regulatory interactions and identified differential signals in cell subtypes.
• Attune revealed chromatin potential for transcription factors and elucidated cell state transitions in differentiation processes.

Conclusions:

• Attune provides a versatile framework for integrating gene expression and chromatin accessibility data.
• The approach facilitates the inference of regulatory mechanisms and gene expression prediction.
• Attune enables advanced omics analyses and a deeper understanding of cellular regulation.