Scholarlink

The rising global incidence of Crohn’s disease (CD) presents significant challenges for disease management. Bifidobacteria demonstrate therapeutic potential through effective gut colonization and alleviation of colitis symptoms, evidenced by reductions in disease activity indices. This study integrated analysis of 28 Bifidobacterium metabolite-related target genes with Crohn’s disease differentially expressed genes (DEGs), identifying 102 relevant genes. Subsequently, 12 machine learning algorithms—including Elastic Net, Gradient Boosting Machine, glmBoost, Lasso, Linear Discriminant Analysis, Naive Bayes, Partial Least Squares Regression, Random Forest, Ridge regression, Stepwise GLM, Support Vector Machine, and XGBoost—were employed in 110 combinatorial approaches for model training and evaluation. The optimal model was selected based on the highest Concordance index (C-index). Integrating protein quantitative trait loci (pQTL)-based Mendelian Randomization (MR) analysis with SHapley Additive exPlanations (SHAP) interpretation identified GRIA4 as a core target gene exhibiting a protective effect against CD.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Bifidobacterium-Crohn’s Disease Therapeutic Target Discovery via Explainable Machine Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Tangyu Yuan,
  • Xuemei Zheng,
  • Jiayin Xing,
  • Zimo Wang,
  • Xinxia Song,
  • Yinghui You,
  • Pengtao Liu

摘要

The rising global incidence of Crohn’s disease (CD) presents significant challenges for disease management. Bifidobacteria demonstrate therapeutic potential through effective gut colonization and alleviation of colitis symptoms, evidenced by reductions in disease activity indices. This study integrated analysis of 28 Bifidobacterium metabolite-related target genes with Crohn’s disease differentially expressed genes (DEGs), identifying 102 relevant genes. Subsequently, 12 machine learning algorithms—including Elastic Net, Gradient Boosting Machine, glmBoost, Lasso, Linear Discriminant Analysis, Naive Bayes, Partial Least Squares Regression, Random Forest, Ridge regression, Stepwise GLM, Support Vector Machine, and XGBoost—were employed in 110 combinatorial approaches for model training and evaluation. The optimal model was selected based on the highest Concordance index (C-index). Integrating protein quantitative trait loci (pQTL)-based Mendelian Randomization (MR) analysis with SHapley Additive exPlanations (SHAP) interpretation identified GRIA4 as a core target gene exhibiting a protective effect against CD.