scGREAT: Transformer-based deep-language model for gene regulatory network inference from single-cell...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Gene regulatory networks (GRNs) are crucial for understanding cellular functions.
• Single-cell sequencing enables GRN inference at a granular level.
• Existing methods face computational costs and simplistic assumptions.

Purpose of the Study:

• To introduce scGREAT, a novel framework for inferring GRNs from single-cell transcriptomics.
• To address computational and assumption-based limitations of current GRN inference tools.

Main Methods:

• scGREAT constructs gene expression and biotext dictionaries from scRNA-seq data.
• It employs a transformer-based engine to learn gene pair representations via optimized embedding spaces.
• The framework integrates gene expression and textual information for robust GRN inference.

Main Results:

• scGREAT demonstrated superior performance compared to contemporary methods on benchmark datasets.
• Gene representations generated by scGREAT offer significant insights into gene regulation.
• External validation using spatial transcriptomics confirmed scGREAT's mechanistic annotations.

Conclusions:

• scGREAT provides an efficient and accurate framework for inferring gene regulatory networks.
• The method offers valuable gene regulatory insights and identifies novel TF-target interactions.
• scGREAT advances the field of single-cell GRN inference.