GatorSC: Multi-Scale Cell and Gene Graphs with Mixture-of-Experts Fusion for Single-Cell Transcriptomics

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high-resolution cellular heterogeneity insights.
• Existing methods underutilize the rich structural information within scRNA-seq data, particularly due to noise and sparsity.
• Integrating heterogeneous graph-based views of cells and genes is crucial for robust low-dimensional representations.

Purpose of the Study:

• Introduce GatorSC, a unified representation learning framework for scRNA-seq data.
• Leverage multi-scale cell and gene graphs for enhanced information fusion.
• Develop noise-robust and structure-preserving embeddings using self-supervised learning.

Main Methods:

• GatorSC models scRNA-seq data using global cell-cell, global gene-gene, and local gene-gene graphs.
• A Mixture-of-Experts architecture adaptively fuses graph neural network embeddings via a gating network.
• A unified self-supervised objective combines graph reconstruction and contrastive learning for both cell and gene graphs.

Main Results:

• GatorSC was evaluated on 19 diverse scRNA-seq datasets.
• Outperformed state-of-the-art methods in cell clustering, gene expression imputation, and cell-type annotation across 14 benchmark datasets.
• Demonstrated accurate trajectory inference and recovery of biological signatures in an Alzheimer's disease dataset.

Conclusions:

• GatorSC provides a flexible and powerful foundation for comprehensive single-cell transcriptomic analysis.
• The framework effectively integrates multi-scale graph structures for robust representation learning.
• GatorSC's approach is extendable to multi-omic and spatial transcriptomic data.