Modeling and analyzing single-cell multimodal data with deep parametric inference.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell multimodal sequencing technologies offer multi-view insights into cellular heterogeneity and disease mechanisms.
• Existing analysis methods may not fully capture the complexity of multimodal single-cell data.

Purpose of the Study:

• To introduce Deep Parametric Inference (DPI), a comprehensive end-to-end framework for single-cell multimodal data analysis.
• To demonstrate DPI's capability in characterizing cellular heterogeneity and analyzing disease progression.

Main Methods:

• DPI transforms single-cell multimodal data into a multimodal parameter space by inferring individual modal parameters.
• The framework was evaluated on cord blood mononuclear cells (CBMC) and peripheral blood mononuclear cells (PBMC) datasets.
• Comparative analyses were performed against state-of-the-art methods.

Main Results:

• DPI's multimodal parameter space provides a more comprehensive characterization of cellular heterogeneity than individual modalities.
• DPI demonstrates superior performance compared to existing methods across multiple datasets.
• DPI effectively analyzes disease progression, such as in COVID-19, and can reference cell types without batch effects.
• A cell state vector field was proposed to analyze cell state transformation patterns.

Conclusions:

• DPI is a powerful framework for single-cell multimodal analysis, offering novel biological insights.
• The framework facilitates a deeper understanding of cellular heterogeneity and disease mechanisms.
• DPI is publicly available with code, datasets, and manuals for biomedical researchers.