Single-cell biological network inference using a heterogeneous graph transformer.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell multi-omics (scMulti-omics) enables simultaneous quantification of multiple molecular layers, offering deep insights into cellular heterogeneity and complex biological mechanisms.
• Current computational tools struggle to effectively infer active biological networks within diverse cell types and their responses to stimuli from scMulti-omics data.

Purpose of the Study:

• To introduce DeepMAPS, a novel computational framework for robust biological network inference from scMulti-omics data.
• To address the limitations of existing methods in capturing cell-type-specific network dynamics and responses to external stimuli.

Main Methods:

• DeepMAPS models scMulti-omics data within a heterogeneous graph structure.
• It employs a multi-head graph transformer to learn intricate relationships among cells and genes in both local and global contexts.
• The method is designed for robust network inference and analysis of complex biological systems.

Main Results:

• Benchmarking demonstrates DeepMAPS surpasses existing tools in cell clustering and biological network construction accuracy.
• DeepMAPS effectively derives cell-type-specific biological networks using lung tumor CITE-seq and lymphoma scRNA-seq/scATAC-seq data.
• Performance evaluation highlights its capability in analyzing diverse scMulti-omics datasets.

Conclusions:

• DeepMAPS provides a powerful and accurate approach for biological network inference from scMulti-omics data.
• The developed DeepMAPS webserver enhances usability and reproducibility for scMulti-omics data analysis, facilitating broader scientific application.