CSGDN: contrastive signed graph diffusion network for predicting crop gene–phenotype associations

• DOI: 10.1093/bib/bbaf062
• Publication Date: 2025-02-20T11:26:28Z
• Authors: Yiru Pan; Xingyu Ji; Jiaqi You; Lu Li; Zhenping Liu; Xianlong Zhang; Zeyu Zhang; Maojun Wang
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Machine Learning (cs.LG)

Abstract Positive and negative association prediction between gene phenotype helps to illustrate the underlying mechanism of complex traits in organisms. The transcription regulation activity specific genes will be adjusted accordingly different cell types, developmental timepoints, physiological states. There are following two problems obtaining positive/negative associations phenotype: (1) high-throughput DNA/RNA sequencing phenotyping expensive time-consuming due need process large sample sizes; (2) experiments introduce both random systematic errors, and, meanwhile, calculations or predictions using software models may produce noise. To address these issues, we propose a Contrastive Signed Graph Diffusion Network, CSGDN, learn robust node representations with fewer training samples achieve higher link accuracy. CSGDN uses signed graph diffusion method uncover regulatory phenotypes. Then, stochastic perturbation strategies used create views for original diffusive graphs. Lastly, multiview contrastive learning paradigm loss is designed unify presentations learned from resist interference reduce We perform validate performance three crop datasets: Gossypium hirsutum, Brassica napus, Triticum turgidum. results show that proposed model outperforms state-of-the-art methods by up 9. 28% AUC sign G. hirsutum dataset. source code our available at https://github.com/Erican-Ji/CSGDN.