面向手性药物布洛芬拆分和运载的高能金属有机框架的机器学习-大数据研究  

A machine Learning-Big Data Study on High-Energy Metal-Organic frameworks for chiral drug Ibuprofen resolution and transportation

作　　者：乔智威 柯诗晴 关珂昕 许方怡 黄晓珊 关雅芳 QIAO Zhi-wei;KE Shi-qing;GUAN Ke-xin;XU Fang-yi;HUANG Xiao-shan;GUAN Ya-fang(School of Chemistry and Chemical Engineering,Guangzhou University,Guangzhou 510006,China)

机构地区：[1]广州大学化学化工学院学院,广东广州510006

出　　处：《广州大学学报（自然科学版）》2024年第5期25-36,共12页Journal of Guangzhou University:Natural Science Edition

基　　金：国家自然科学基金项目(22478085,21978058);广东省自然科学基金项目(2023A1515240076,2022A1515011446)。

摘　　要：药物分离/药物负载材料已成为药物可控释放及药物制备技术中的重要研究对象之一。为了从大量的已有材料中选取可高效负载药物的候选者并探索其负载机理,该研究从CoRE-MOF 2019数据库中抽取了1 000种MOFs材料,通过高通量计算探讨它们对布洛芬药物的吸附负载性能。首先,将MOFs的8种结构/能量描述符分别与MOFs对布洛芬药物分子的选择性(S_(S-IBU/N2))、吸附量(N_(S-IBU))和两者权衡值(Trade-off between S_(S-IBU/N2)and N_(S-IBU))进行单变量分析,初步探索了不同描述符和3种性能指标之间的关系趋势。其次,采用随机森林、极限梯度提升(Extreme Gradient Boosting, XGB)、梯度提升树、轻量梯度提升树、反向传播神经网络和支持向量回归6种机器学习算法,分别对8种描述符和3种性能评价标准(吸附选择性S_(S-IBU/N2)、吸附量N_(S-IBU)和权衡值TSN)进行大数据训练和挖掘,并建立定量关系。结果表明,6种ML算法的预测精度都是N_(S-IBU)>TSN>S_(S-IBU/N2)。对于S_(S-IBU/N2)来说,XGB表现出最佳的预测效果(R^(2)=0.83)。接着,基于XGB模型使用形状添加解释器(SHapley Additive explanation, SHAP)方法来解释和分析MOF描述符对性能指标的重要程度。MOF吸附过程中产生的总能量被认为是关键的影响因素,它与TSN和N_(S-IBU)都呈现正相关的趋势。最终,结合毒理学分析,推荐和设计了一系列高性能MOF材料。文章从分子层面、高通量计算到大数据挖掘,系统地研究了布洛芬药物分子在MOF中的吸附运载机理,为药物运载材料提供了理论指导。Drug separation/drug loading materials have become one of the significant research objects in controlled drug release and drug preparation technology.In order to select candidates for efficient drug loading from a large number of existing materials and explore their loading mechanisms,1000 MOFs materials were extracted from the CoRE-MOF 2019 database for this study,and their adsorptive loading performance the drug ibuprofen was explored by high-throughput calculations.Firstly,the eight structure/energy descriptors of MOFs were analyzed by univariate analysis with the adsorption se-lectivity(S_(S-IBU/N2)),adsorption capacity(N_(S-IBU))and trade-off value(TSN)of MOFs for ibuprofen drug molecules,and the relationship trend between different descriptors and the three performance in-dicators were preliminarily explored.Secondly,six machine learning models including Random For-est,Extreme Gradient Boosting(XGB),Gradient Boosting Decision Tree,Light Gradient Boosting Machine,Backpropagation Neural Network,and Support Vector Regression six machine learning algo-rithms with eight descriptors and three performance evaluation criteria(Adsorption selectivity S_(S-IBU/N2),Adsorption capacity N_(S-IBU/N2) and Trade-off value TSN)for big data training and mining,were used to establish quantitative relationships.The results show that the prediction accuracy of the six ML algo-rithms is N S-IBU>TSN>S_(S-IBU/N2).For S_(S-IBU/N2),XGB showed the best prediction(R^(2)=0.83).Subse-quently,based on the XGB model,the SHaple Additive explanation(SHAP)method was used to ex-plain and analyze the importance of MOF descriptors to performance indicators.The total energy gen-erated during MOF adsorption is considered to be the key influencing factor,and it shows a positive correlation trend with both TSN and N S-IBU.Finally,combined with toxicological analysis,a series of high-performance MOF materials were recommended and designed.This work,from molecular level,high-throughput computing to big data mining,systematically studied the adsorption

关 键 词：布洛芬 金属有机框架 机器学习 

分 类 号：R979.9[医药卫生—药品]