Inferring cell-type-specific gene regulatory network from cellular transcriptomics data with GeneLink.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Cell-type-specific gene regulatory networks (ctGRNs) are vital for understanding biological processes like development and cancer.
• Inferring ctGRNs from transcriptomic data is challenging due to data sparsity, cell heterogeneity, and over-smoothing in deep learning models.

Purpose of the Study:

• To present GeneLink+, an advanced framework for ctGRN inference using directed graph link prediction.
• To enhance the accuracy and interpretability of ctGRN inference by addressing limitations of existing deep learning approaches.

Main Methods:

• GeneLink+ employs residual-GATv2 blocks, combining attention mechanisms and residual connections to preserve gene features and mitigate information loss.
• A modified dot product scheme with learnable weights adaptively prioritizes gene pairs for precise causal edge attribution.

Main Results:

• GeneLink+ outperforms or matches state-of-the-art methods in predictive accuracy and biological relevance across seven benchmark datasets.
• The framework successfully identified key regulatory relationships in transcriptomic data from blood immune cells, Alzheimer's disease, and breast cancer.

Conclusions:

• GeneLink+ offers a robust and interpretable solution for ctGRN inference from diverse transcriptomic data.
• The framework advances the study of gene regulation in complex biological systems and diseases.