GLUE

GLUE generates graph-linked unified embeddings to integrate single-cell multi-omics datasets and align disparate feature spaces using prior regulatory interaction knowledge.

Key Features:

• Graph-Linked Unified Embedding: Generates unified embeddings that harmonize data across multiple single-cell omics layers using a graph-linked approach.
• Unpaired multi-omics integration: Aligns and integrates unpaired single-cell multi-omics datasets by bridging feature discrepancies across different omics layers.
• Utilization of regulatory interactions: Incorporates prior knowledge of regulatory interactions to guide alignment of features between omics modalities.
• Scalability and Robustness: Demonstrated accurate, scalable, and robust performance on large-scale datasets in systematic benchmarks.
• Versatile Applications: Applicable to triple-omics integration, model-based regulatory inference, and construction of multi-omics human cell atlases from millions of cells.
• Modular and Extensible Design: Implements a modular framework that supports extension to additional analysis tasks.

Scientific Applications:

• Unpaired multi-omics dataset integration: Enables integration of unpaired datasets across modalities to produce joint embeddings for downstream analysis.
• Triple-omics integration: Combines three omics layers in a single unified embedding to analyze multi-modal cellular states.
• Model-based regulatory inference: Supports inference of regulatory interactions and gene regulation mechanisms using model-based approaches.
• Multi-omics human cell atlas construction: Scales to integrate data from millions of cells to support construction of comprehensive multi-omics human cell atlases.

Methodology:

Constructs graph-linked unified embeddings and aligns features across omics layers by incorporating prior regulatory interaction knowledge.