Risk prediction of Helicobacter pylori strains across Correa’s cascade via intelligent analysis of genome-wide SNPs
- Xiuling Song
- Xiangyu Wang
- Shuzhen Zhang
- Hong Li
- Chao Wu
- Huifang Liu
- Chin Yen Tay
- Cong Ma
- Barry J. Marshall
- Bing Gu
- Liang Wang
2026-06-09
Only a minority of Helicobacter pylori ( H. pylori )-infected individuals progress along Correa’s cascade, and classical virulence markers do not fully explain this heterogeneity, motivating genome-wide approaches to quantify strain-level genomic risk associated with advanced lesions. We assembled 528 high-quality H. pylori whole-genome sequences spanning nonatrophic gastritis (NAG), atrophic gastritis (AG), intestinal metaplasia (IM), and gastric cancer (GC) and performed bacterial genome-wide association analyses with explicit adjustment for population structure. We then integrated advanced lesions-associated variants into a random forest model to derive an H. pylori Genomic Risk Score (HpRS), defined as the predicted probability that a strain is associated with advanced lesions (IM/GC) vs. nonadvanced lesions (NAG/AG). Despite phylogenetic analyses revealing that genome-wide clustering was primarily driven by geographic lineage rather than disease stage, HpRS achieved strong discrimination in repeated cross-validation (mean AUC = 0.902, 95% CI: 0.892 to 0.912), remained discriminatory in an internal held-out set (AUC = 0.780), and generalized to two independent cohorts (Bacterial and Viral Bioinformatics Resource Center (BV-BRC): AUC = 0.871; hospital cohort distinguishing IM vs. NAG/AG: AUC = 0.843). Predictive single-nucleotide polymorphisms mapped mainly to core functions (DNA repair, translation, and central and lipid metabolism) and often affected conserved domains, suggesting a polygenic architecture and generating testable functional hypotheses. HpRS provides a proof-of-principle framework for strain-aware risk stratification and may complement future gastric cancer prevention strategies.