This work proposed a novel robust framework, HybridGRN, for gene regulatory network inference based on local topology and dynamic thresholds. The key innovations include: (1) combine local topology perturbations with sample perturbations to reduce false positives; (2) introduce a dynamically adjustable threshold based on target gene connectivity to suppress redundant edges in hub genes; (3) iteratively delete nodes and retrain subnetworks, enhancing the detection of reverse edges and feedback loops. HybridGRN exhibits remarkable potential and computational efficiency even in small-sample scenarios, expanding its applicability in analyzing complex biological processes.
Li et al. (Thu,) studied this question.