Inferring Gene Regulatory Networks From Single-Cell RNA Sequencing Data by Dual-Role Graph Contrastive Learning.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Gene regulatory network (GRN) inference is crucial for understanding cellular mechanisms.
• Existing methods struggle with single-cell RNA sequencing data due to noise and sparsity.
• Accurate cell-type-specific GRNs are needed for insights into cellular identity and disease.

Purpose of the Study:

• To present RegGAIN, a novel deep learning model for inferring gene regulatory networks from single-cell transcriptomic data.
• To improve the accuracy and robustness of GRN reconstruction.
• To enable the discovery of condition-specific and dynamic regulatory programs.

Main Methods:

• RegGAIN utilizes self-supervised contrastive learning to enhance gene embeddings.
• Dual-role representations are learned using separate encoders for directionality and distinct patterns.
• Model performance is evaluated against existing GRN inference methods.

Main Results:

• RegGAIN consistently outperforms current methods in GRN reconstruction accuracy and robustness.
• Predicted regulatory interactions are validated using external epigenetic data.
• The model successfully identifies GRN rewiring and dynamic regulatory programs.

Conclusions:

• RegGAIN provides a powerful and generalizable framework for gene regulatory network inference.
• The model offers deeper insights into cellular regulation across various biological contexts.
• RegGAIN advances the field of computational biology for transcriptomic data analysis.