A multi-view latent variable model reveals cellular heterogeneity in complex tissues for paired multimodal...

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell multimodal assays offer deep insights into cellular heterogeneity and disease mechanisms.
• Existing methods struggle with accurate dimensionality reduction for joint-modality data, limiting the discovery of rare cell subpopulations.

Purpose of the Study:

• To develop a computational framework for accurate integration of paired multimodal single-cell data.
• To improve the identification of novel or rare cell subtypes and facilitate cell lineage inference.

Main Methods:

• VIMCCA (Variational-assisted Multi-view Canonical Correlation Analysis) framework utilizing variational inference and deep learning.
• Jointly learning an inference model and two modality-specific non-linear models.
• Comparison with 10 state-of-the-art algorithms on four diverse multimodal datasets.

Main Results:

• VIMCCA effectively integrates various joint-modality data types, leading to more reliable downstream analyses.
• Demonstrated superior performance in identifying novel or rare cell subtypes compared to existing methods.
• Facilitated cell lineage inference based on joint-modality profiles.

Conclusions:

• VIMCCA provides a robust and accurate computational framework for multimodal single-cell data integration.
• The framework enhances the discovery of cellular heterogeneity and improves understanding of cell development and disease.