Work

Dual-task graph attention learning for stress-responsive gene regulatory network inference in Panax notoginseng

Year: 2026
Type: article
Abstract: Panax notoginseng is an important specialty crop and geo-authentic medicinal plant vulnerable to multiple stresses, with its stress responses regulated by complex gene regulatory networks. Limited species-specific regulatory evidence and sample size make systematic analysis of transcription factor (TF)-target relationships difficult. We developed a dual-task graph attention network (DTGAT) with a shared graph attention encoder and joint optimization of link prediction and expression reconstruction for regulatory relationship prediction. DTGAT was first evaluated in Arabidopsis thaliana for methodological validation. On the independent test set, link prediction achieved an AUC of 0.9106 and an AUPR of 0.8933, while expression reconstruction achieved an R² of 0.8375 and a PCC of 0.9152; link prediction outperformed the evaluated baselines. DTGAT was then applied to transcriptomic data from stressed P. notoginseng together with a homology-mapped prior network, yielding 55,108 high-confidence candidate regulatory edges. Promoter motif analysis further provided sequence-level support for 22 newly predicted candidate regulatory relationships associated with stress responses. Published experimental evidence supported the biological plausibility of the candidate MYB–CCoAOMT relationship. These findings highlight the utility of DTGAT for prioritizing stress-responsive regulatory candidates in non-model medicinal plants.
Language: English
Open Access status: gold
Estimated APC paid: $3,801
OpenAlex ID: W7235003937
FWCI:
Cites: 38
Cited by:
Related to: 10
Parts: 9