AutoGERN: single-cell RNA-seq gene regulatory network inference via explicit link modeling and adaptive...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution transcriptome data, revealing cellular heterogeneity.
• Inferring gene regulatory networks (GRNs) from scRNA-seq is crucial but challenging due to complex regulatory dependencies.
• Existing graph neural network (GNN) methods often implicitly model regulatory relationships, limiting their effectiveness.

Purpose of the Study:

• To develop an advanced GNN framework, AutoGERN, for robust GRN inference from scRNA-seq data.
• To explicitly model regulatory information and learn expressive link embeddings for improved gene-gene association scoring.
• To enhance model flexibility and generalization across diverse scRNA-seq datasets.

Main Methods:

• AutoGERN utilizes dual message-passing spaces (within-layer and cross-layer) for enhanced representational power.
• It explicitly models regulatory information within the message-passing framework.
• An AutoGNN-based architecture search is integrated to adapt the network to different data distributions.

Main Results:

• AutoGERN achieves superior performance in GRN inference compared to state-of-the-art methods.
• The framework demonstrates enhanced robustness across multiple real-world scRNA-seq datasets.
• Explicit modeling of regulatory information leads to more accurate gene-gene regulatory association scoring.

Conclusions:

• AutoGERN offers a flexible and powerful GNN framework for accurate GRN inference from scRNA-seq data.
• Its adaptive architecture search improves generalization across datasets with varying distributions.
• The explicit modeling of regulatory dependencies represents a significant advancement in the field.