MIRA

MIRA integrates single-cell gene expression and chromatin accessibility data using probabilistic multimodal models to infer regulatory potential and cell-state trajectories.

Key Features:

• Probabilistic Multimodal Modeling: Employs probabilistic models to compare transcriptional activity and chromatin accessibility at single-cell resolution.
• Topic Modeling of Cell States: Uses topic modeling to represent cell states in an interpretable latent space for identifying distinct state trajectories.
• Regulatory Potential Modeling: Models regulatory potential at individual gene loci to reveal locus-specific regulatory influences on transcription.
• High-Fidelity Cell State Trees: Infers hierarchical trees of cell states that depict differentiation paths and branch points.
• Variable Influence Analysis: Assesses the locus-specific influence of local chromatin accessibility on transcription across genes and developmental stages.

Scientific Applications:

• Epidermal differentiation: Applied to epidermal differentiation, demonstrating that early developmental genes are tightly regulated by local chromatin landscapes while terminal fate genes require less chromatin remodeling.
• Embryonic brain development: Applied to embryonic brain development, revealing differential chromatin regulation of early versus terminal fate genes during development.

Methodology:

Implements probabilistic multimodal modeling, topic modeling for latent cell-state representation, regulatory-potential modeling at gene loci, inference of cell-state trees, and locus-specific analysis of chromatin accessibility influence on transcription.