scGraph2Vec: a deep generative model for gene embedding augmented by graph neural network and single-cell omics data.

Area of Science:

• Computational biology
• Genomics
• Systems biology

Background:

• Understanding gene function in complex diseases requires analyzing biological networks.
• Gene-gene interaction networks offer insights into cellular processes and disease mechanisms.

Purpose of the Study:

• To develop a deep learning framework for generating informative gene embeddings.
• To leverage single-cell data and gene-gene interaction networks for biological insights.

Main Methods:

• Implemented scGraph2Vec, a variational graph autoencoder framework.
• Integrated single-cell datasets with gene-gene interaction networks.
• Generated biologically interpretable gene embeddings.

Main Results:

• scGraph2Vec embeddings identified functional and tissue-specific gene clusters.
• The framework outperformed existing tools in distinguishing gene clusters and functional aggregation.
• Applied embeddings to infer disease-associated genes (e.g., COVID-19, Alzheimer's), identify lung adenocarcinoma driver genes, and reveal melanoma cell state regulators.

Conclusions:

• scGraph2Vec reconstructs tissue-specific gene networks.
• The generated latent gene representations imply biological functions and aid disease gene discovery.