VI-VS

VI-VS balances flexibility and interpretability to identify relevant feature relationships in single-cell multiomic data using deep generative models and variational inference for variable selection with false discovery rate control.

Key Features:

• Calibrated Variable Selection: Uses variational inference for variable selection to identify features more likely representing genuine causal relationships.
• False Discovery Rate Control: Provides statistical control over false discoveries when identifying conditionally dependent features.
• Deep Generative Models: Employs flexible deep generative models to capture complex dependencies while maintaining interpretability.
• Conditional Dependency Detection: Identifies conditionally dependent features across multiomic data types.

Scientific Applications:

• Single-Cell Multiomics Analysis: Unveils functional relationships between various molecular cell phenotypes from single-cell multiomic data.
• Causal Feature Identification: Distinguishes features more likely representing causal relationships from merely correlated features.
• Multiomic Integration: Integrates diverse molecular phenotype data to identify cross-modality feature dependencies.

Methodology:

VI-VS uses deep generative models with variational inference for variable selection.