Data integration and inference of gene regulation using single-cell temporal multimodal data with scTIE

Area of Science:

• Computational Biology
• Genomics
• Developmental Biology

Background:

• Single-cell technologies enable detailed analysis of gene regulation.
• Integrating multimodal single-cell data (scRNA-seq, scATAC-seq) is crucial for cell type identification but remains challenging.
• Existing methods often treat data integration separately from regulatory relationship inference.

Purpose of the Study:

• To present scTIE, a unified computational method for integrating temporal multimodal single-cell data.
• To infer gene regulatory relationships predictive of cellular state transitions.
• To improve cell type identification and understanding of developmental processes.

Main Methods:

• scTIE employs an autoencoder with iterative optimal transport for embedding cells across time points.
• It extracts interpretable information to predict cell trajectories.
• The method was validated on synthetic and real temporal multimodal datasets, including mouse embryonic stem cell differentiation.

Main Results:

• scTIE effectively integrates temporal multimodal data, outperforming existing methods in preserving biological signals, especially with batch effects and noise.
• The method successfully identifies regulatory elements predictive of cell transition probabilities.
• It demonstrates robust performance in capturing dynamic biological processes.

Conclusions:

• scTIE offers a unified framework for temporal multimodal single-cell data integration and regulatory network inference.
• This approach enhances the understanding of regulatory landscapes driving developmental processes.
• scTIE provides a powerful tool for dissecting context-specific gene regulatory mechanisms.