MESSI

MESSI identifies signaling genes active in cell-cell interactions within spatial single-cell expression datasets to predict response gene levels and characterize cellular communication.

Key Features:

• Mixture of Experts Strategy: Employs a mixture of experts approach to subdivide cells into distinct subtypes for modeling heterogeneity.
• Multi-task Learning: Uses multi-task learning to incorporate information from neighboring cells for joint prediction of response genes.
• Spatial Context Integration: Integrates spatial context and neighboring-cell information to model inter-cellular signaling.
• Joint Intra- and Inter-cellular Modeling: Jointly models gene interactions within individual cells and between neighboring cells to infer active signaling.
• Response Gene Prediction: Predicts response gene levels based on inferred signaling both intra-cellularly and inter-cellularly.

Scientific Applications:

• Spatial transcriptomics analysis: Applied to spatial transcriptomics data to study cellular communication at single-cell resolution.
• Prediction of response genes: Predicts response gene levels to reveal signaling pathways and characterize cell subtypes.
• Neuroscience applications: Used to analyze subtypes of excitatory neuron cells to reveal signaling dynamics relevant to neural function and disorders.
• Benchmarking on datasets: Applied to three spatial single-cell expression datasets and reported improved prediction of response gene levels over prior methods.

Methodology:

MESSI uses a mixture of experts strategy and multi-task learning to jointly model gene interactions within cells and between neighboring cells, integrating spatial context and cell type subtyping.