Yapay Zeka Hızlı Bakış & Çıkarımlar
AI InsightThis integrative framework provides a transparent and reproducible approach for gene discovery in scRNA-seq data and contributes computationally prioritized candidate genes warranting experimental validation for understanding MALT lymphoma pathogenesis and improving future molecular diagnostics.
📌 3 Maddede Bu Makale
Özet
Mucosa-associated lymphoid tissue (MALT) lymphoma is an indolent and clinically heterogeneous B-cell malignancy. The lack of robust molecular biomarkers and the limited availability of high-resolution single-cell studies hinder the development of precise diagnostic and therapeutic strategies. To address this gap, an interpretable machine learning pipeline was developed and applied to single-cell RNA sequencing (scRNA-seq) data derived from MALT lymphoma patients and healthy donors. After quality control, normalization, and dimensionality reduction via truncated singular value decomposition (SVD), multiple tree-based classifiers were trained using stratified cross-validation and evaluated on a held-out balanced test set. Feature importance scores and SHapley Additive exPlanations (SHAP) values were integrated with Wilcoxon rank-sum testing to identify statistically supported gene markers. Among the classifiers, CatBoost, LightGBM, and Random Forest achieved the highest performance. Genes such as RPS4Y1, RGS1, XIST, CREM, and HSPH1 were consistently prioritized by both SHAP and statistical testing, indicating their biological relevance and differential expression. Notably, a divergence was observed between impurity-based feature importance rankings and the SHAP/Wilcoxon consensus, reflecting the complementary nature of these analytical approaches. This integrative framework provides a transparent and reproducible approach for gene discovery in scRNA-seq data and contributes computationally prioritized candidate genes warranting experimental validation for understanding MALT lymphoma pathogenesis and improving future molecular diagnostics.
Abstract
Mucosa-associated lymphoid tissue (MALT) lymphoma is an indolent and clinically heterogeneous B-cell malignancy. The lack of robust molecular biomarkers and the limited availability of high-resolution single-cell studies hinder the development of precise diagnostic and therapeutic strategies. To address this gap, an interpretable machine learning pipeline was developed and applied to single-cell RNA sequencing (scRNA-seq) data derived from MALT lymphoma patients and healthy donors. After quality control, normalization, and dimensionality reduction via truncated singular value decomposition (SVD), multiple tree-based classifiers were trained using stratified cross-validation and evaluated on a held-out balanced test set. Feature importance scores and SHapley Additive exPlanations (SHAP) values were integrated with Wilcoxon rank-sum testing to identify statistically supported gene markers. Among the classifiers, CatBoost, LightGBM, and Random Forest achieved the highest performance. Genes such as RPS4Y1, RGS1, XIST, CREM, and HSPH1 were consistently prioritized by both SHAP and statistical testing, indicating their biological relevance and differential expression. Notably, a divergence was observed between impurity-based feature importance rankings and the SHAP/Wilcoxon consensus, reflecting the complementary nature of these analytical approaches. This integrative framework provides a transparent and reproducible approach for gene discovery in scRNA-seq data and contributes computationally prioritized candidate genes warranting experimental validation for understanding MALT lymphoma pathogenesis and improving future molecular diagnostics.