A microbiota-based perspective on urinary stone disease: Insights from 16S rRNA sequencing and machine learning models

Yufeng LiuAoyu YangZiyi ZhangChen ShenWei WangXiancheng Li^*

• Department of Urology, The Second Affiliated Hospital of Dalian Medical University, Dalian, China

Background: Urinary stones are a multifactorial disease. Recently, the role of microorganisms in pathogenesis has attracted attention. Although studies suggested certain microbes in gut and urine are associated with urinary stones, current classification criteria remain insufficient. This study analyzed gut and urinary microbiota via 16S rRNA sequencing in patients with pure CaOx, UA, and Inf stones. By integrating microbiota with clinical data, we constructed machine learning models to evaluate their diagnostic value. Methods: A total of 81 urinary stone patients (30 CaOx, 31 UA, 20 Inf) and 26 healthy volunteers were enrolled. Stool and urine samples underwent 16S rRNA sequencing to characterize microbiota. We integrated clinical data (age, gender, BMI) using LASSO feature selection and six algorithms (SVM, Random Forest, XGBoost, etc.) to create predictive models. Model performance was evaluated by cross-validation. Results: Enrichment of Paramuribaculum, Muribaculum, Mesorhizobium, and Acinetobacter was found in the gut of CaOx patients, with urinary enrichment of Enterococcus. UA stone patients showed increased gut Massilioclostridium and urinary Fenollaria, Anaerococcus, Enterococcus, and Escherichia. Inf stone patients showed no gut differences but exhibited urinary enrichment of Escherichia. The predictive model combining urinary microbiota and clinical data performed excellently (AUC: SVM 0.922, Random Forest 0.866, XGBoost 0.913). Conclusion: Different stone types are characterized by distinct microbiota. Integrating microbiota with machine learning enables non-invasive prediction of stone types, providing insights into microecological mechanisms and new avenues for diagnosis.