ScADSATGRN: An Adaptive Diffusion Structure-Aware Transformer Based Method Inferring Gene Regulatory Networks from Single-Cell Transcriptomic Data

Gene regulatory networks unveil the interactions and regulatory relationships between genes, offering profound insights into cellular functional mechanisms. Utilizing single-cell RNA sequencing (scRNA-seq) data, we can take advantage of unprecedented opportunities to reconstruct gene regulatory networks (GRNs) at ultra-fine resolution, thereby uncovering intricate details of gene regulation. However, the current accuracy of using single-cell transcriptome data to infer gene regulatory networks needs to be improved. Therefore, in this article, we introduce the Transformer concept into the inference of gene regulatory networks. We propose a graph neural network model based on the Transformer architecture. The model combines GNN with Transformers to learn graph-structured data, enabling it to capture global within graph information. Compared with several existing methods, our model demonstrates superior performance across seven scRNA-seq datasets containing four types of ground truth networks. This facilitates the study of gene regulatory networks.