可解释AI预测植物化合物肠道吸收特性
Explainable AI predicts how plant compounds may cross intestinal barriers
韩国科学家开发了可解释的机器学习模型,直接从分子结构预测植物化合物的肠道吸收特性,无需实验。该模型在预测渗透性和外排比方面表现优异,有助于筛选功能性食品和新药候选物。
Phytochemicals are bioactive compounds derived from plants, such as flavonoids, alkaloids and ginsenosides. They are key ingredients in functional foods and promising starting points for new drugs. Yet many of them never make it into the bloodstream because they are poorly absorbed in the intestine. Measuring their absorption one compound at a time requires laborious cell-based experiments, and this bottleneck has long slowed the screening of candidate ingredients.
Researchers at Sejong University and the Korea Institute of Science and Technology (KIST) have now developed explainable machine learning models that predict key intestinal absorption properties of phytochemicals directly from their molecular structure, before any laboratory experiment is performed. The study was published in the Journal of the Science of Food and Agriculture.
The team focused on three indicators that are routinely measured in the human intestinal Caco-2 cell model, the standard laboratory system for studying intestinal absorption. Apparent permeability (Papp) reflects how readily a compound crosses the intestinal cell layer. Transepithelial electrical resistance (TEER) indicates the integrity of the intestinal barrier. The efflux ratio (ER) reveals whether transporter proteins actively pump a compound back out of the cells.
Starting from experimentally measured values for 83 structurally diverse phytochemicals, the researchers computed 5,003 molecular descriptors from each chemical structure. These descriptors are numerical features that encode the electronic, topological and structural characteristics of a compound.
A workflow combining statistical feature selection, recursive feature elimination and repeated cross-validation was used to compare 10 regression algorithms for each indicator.
The best performance came from a stacking ensemble model that combines CatBoost, LightGBM and Gradient Boosting as base learners with a linear regression meta-model. The ensemble achieved strong predictive accuracy for apparent permeability and efflux ratio, while TEER proved harder to predict and showed moderate but more variable performance.
In the model screening stage, cross-validated models reached coefficients of determination of up to about 0.96 for apparent permeability. The team also built a classification model that sorts compounds by efflux behavior, and it achieved an area under the curve of 0.96 in cross-validation.
A distinctive feature of the study is its emphasis on interpretability. Using SHAP (SHapley Additive exPlanations), an explainable artificial intelligence technique, the researchers identified which molecular features drive each prediction. Electronic, topological and branching-related descriptors emerged as key determinants of absorption.
As a result, the models do not simply output a number. They also show why a compound is predicted to be well or poorly absorbed, offering chemists concrete clues for designing better-absorbed ingredients.
The authors suggest that such explainable QSPR (quantitative structure-property relationship) models can serve as a practical prescreening tool in the discovery of functional food ingredients and drug candidates. Researchers could prioritize compounds that are likely to be bioavailable and reserve costly laboratory assays for the most promising ones.
The work was carried out by the laboratory of Professor Soon-Mi Shim in the Department of Food Science and Biotechnology at Sejong University, in collaboration with the KIST Gangneung Institute of Natural Products, with Jin-Woo Kim as first author.
Publication details
Jin‐Woo Kim et al., Molecular descriptor driven QSPR modeling of Papp, TEER and Efflux Ratio from Caco‐2 cells using machine learning for various phytochemicals, Journal of the Science of Food and Agriculture (2026). DOI: 10.1002/jsfa.70701
Journal information: Journal of the Science of Food and Agriculture
Key medical concepts
Who's behind this story?
Sadie Harley
BSc Life Sciences & Ecology. Microbiology lab background with pharmaceutical news experience in oil, gas, and renewable industries. Full profile →
Andrew Zinin
Master's in physics with research experience. Long-time science news enthusiast. Plays key role in Science X's editorial success. Full profile →
Citation: Explainable AI predicts how plant compounds may cross intestinal barriers (2026, November 6) retrieved 6 November 2026 from https://medicalxpress.com/news/2026-11-ai-compounds-intestinal-barriers.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.
来源:Medical Xpress · medicalxpress.com