TDAGENE: Inference of Gene Regulatory Network Based on Topological Data Analysis and Graph Attention Network for...

The emergence of scRNA-seq has enabled high-resolution gene expression analysis at the single-cell level, providing important opportunities for inferring gene regulatory networks (GRNs) within individual cells. This study proposes a novel method, termed Topological Data Analysis-guided Gene Network Embedding (TDAGENE), which introduces the topological data analysis (TDA) to enhance the GRN inference. It integrates the global topological feature with local graph representation and, therefore, improves its ability to model the gene expression by capturing the topological structure of GRN and facilitate the identification of gene interaction relationships. Various experiments demonstrate that TDAGENE outperforms existing methods in GRN inference tasks. It achieves optimal predictions on 90% of datasets in terms of area under the precision-recall curve (AUPRC) and optimal performance on 66.7% of datasets in terms of area under the receiver operating characteristic curve (AUROC). Compared to the latest methods, it shows an average improvement of 17.66% in AUPRC and 3.08% in AUROC. Additionally, we apply TDAGENE to analyze 3 key regulators (NANOG, SOX2, and POU5F1), revealing that incorporating topological information effectively captures critical features during cell fate specification. These findings highlight the potential of TDAGENE in inferring GRNs.