A Robust Deep Temporal Causal Discovery Platform for Single‐Cell Gene Regulatory Network Reconstruction

Accurately inferring gene regulatory networks (GRNs) from single‐cell RNA sequencing (scRNA‐seq) data is critical for understanding cellular dynamics in both normal development anddisease. However, existing computational methods often suffer from low precision and high false‐positive rates due to the intrinsic noise and complex regulatory architecture in scRNA‐seq data. We introduce scTIGER2.0, a deep‐learning‐based framework that integrates expression correlation, pseudotime ordering, temporal causal discovery, and bootstrap‐based significance testing to infer high‐confidence, directional gene–gene interactions. Benchmarking against five popular GRN inference methods using large‐scale datasets, scTIGER2.0 consistently achieved superior specificity, especially in linear developmental trajectories. In real applications, scTIGER2.0 identified an APOE ‐centered GRN from Alzheimer's disease scRNA‐seq data and uncovered interconnected GRNs for FOS , FOXP1 , JUN , KLF6 , NCOA4 , and RUNX1 from acute myeloid leukemia data, where 87.5% of the predicted targets show promoter‐binding peaks in the corresponding ChIP‐seq data. These results demonstrate that scTIGER2.0 is a robust, accurate and fully integrated platform for uncovering biologically meaningful GRNs from noisy scRNA‐seq data.