可解释性
数量结构-活动关系
人工智能
机器学习
分子描述符
特征选择
梯度升压
随机森林
线性回归
马修斯相关系数
回归
工作流程
计算机科学
Boosting(机器学习)
回归分析
化学
相关系数
预测建模
化学信息学
偏最小二乘回归
生物系统
支持向量机
数学
交叉验证
模式识别(心理学)
特征(语言学)
相关性
数据挖掘
堆积
膜透性
生物利用度
作者
Jin‐Woo Kim,Rixing Cong,Jin‐Soo Park,Keunwan Park,Kyungsu Kang,Soon‐Mi Shim
摘要
BACKGROUND: The present study aimed to develop and validate quantitative structure-property relationship (QSPR) models for predicting permeability related bioavailability indicators including apparent permeability (Papp), trans-epithelial electrical resistance (TEER) and efflux ratio (ER) based on molecular descriptors (n = 5003) of 83 phytochemicals. A rigorous workflow integrating feature selection and 10 regression algorithms were employed to assess predictive performance. RESULTS: Pearson correlation coefficient analysis revealed significant relationships among the three indictors. Among the models, a stacking ensemble, including CatBoost, LightGBM and Gradient Boosting as base learners and linear regression as the meta-model, achieved strong predictive accuracy for Papp and ER, at the same time as showing moderate but highly variable performance for TEER. SHapley Additive exPlanations (i.e. SHAP)-based feature importance analysis provided insight into key molecular descriptors associated with electronic, topological and branching. CONCLUSION: These results demonstrate the utility of interpretability tools in constructing robust and explainable QSPR models for bioavailability prediction in various phytochemicals. © 2026 The Author(s). Journal of the Science of Food and Agriculture published by John Wiley & Sons Ltd on behalf of Society of Chemical Industry.
科研通智能强力驱动
Strongly Powered by AbleSci AI