Matilda

Matilda implements a multi-task learning framework for integrative analysis of multimodal single-cell omics data, enabling simultaneous data simulation, dimensionality reduction, cell type classification, and feature selection across modalities.

Key Features:

• Multi-task learning framework: Employs a multi-task learning approach that leverages relationships among analytical tasks.
• Supported tasks: Performs data simulation, dimensionality reduction, cell type classification, and feature selection within a unified model.
• Multimodal integration: Integrates information across multiple single-cell omics modalities for joint analysis.
• Task interdependency exploitation: Uses interdependencies between tasks to improve learning outcomes across objectives.
• Implementation: Implemented using the PyTorch deep learning library.

Scientific Applications:

• Integrative analysis of multimodal single-cell omics: Enables joint analysis of datasets containing multiple molecular modalities at single-cell resolution.
• Cell type classification: Facilitates identification and classification of cell types from multimodal single-cell data.
• Data simulation: Produces simulated multimodal single-cell omics data for method evaluation and benchmarking.
• Dimensionality reduction: Provides reduced representations for visualization and downstream analyses of multimodal data.
• Feature selection: Identifies informative features and modality-specific markers across modalities.

Methodology:

Matilda applies a multi-task learning approach that jointly models data simulation, dimensionality reduction, cell type classification, and feature selection by leveraging task interdependencies, and is implemented in PyTorch.