Machine Learning-Enabled Repurposing and Design of Antifouling Polymer Brushes
Machine Learning-Enabled Repurposing and Design of Antifouling Polymer Brushes
复制标题
DOI：
10.1016/j.cej.2021.129872
复制
发表时间：
2021-04
影响因子：
15.1
通讯作者：
Yonglan Liu;Dong Zhang;Yijing Tang;Yanxian Zhang;Xiong Gong;Shaowen Xie;Jie Zheng
中科院分区：
文献类型：
--
作者：
Yonglan Liu;Dong Zhang;Yijing Tang;Yanxian Zhang;Xiong Gong;Shaowen Xie;Jie Zheng
合理开发高分子材料对基础研究和实际应用具有重要意义。然而，由于聚合物及其相关结构/性质的数据复杂性，目前的实验设计和计算建模的聚合物材料仍然保留经验风味。在这项工作中，我们开发了一个数据驱动的机器学习工作流程，结合内部基准数据集的聚合物刷，发现现有的聚合物刷的潜在性能使用基于神经网络的人工神经网络（ANN）模型和设计新的聚合物刷使用基于组的支持向量回归（SVR）模型。由此产生的两个机器学习模型不仅证明了它们的可靠性，预测性和适用性，而且还建立了使用描述符和官能团的组成-结构-性质关系。最后，我们合成了不同的改性和新设计的聚合物刷，预测人工神经网络和SVR模型，所有这些都表现出优异的表面电阻，从未稀释的人血清和血浆中的蛋白质吸附在最佳的膜厚度。总的来说，我们的数据驱动的机器学习模型可用作确定、重新利用和设计聚合物刷以外的新型上级可编程材料的智能工具。
Rational development of antifouling materials is of great importance for fundamental research and real-world applications. However, current experimental designs and computational modelings of antifouling materials still retain empirical flavor due to the data complexity of polymers and their associated structures/properties. In this work, we developed a data-driven, machine learning workflow, in combination with an in-house benchmark dataset of antifouling polymer brushes, to discover the potential antifouling property of existing polymer brushes using the descriptor-based artificial neural network (ANN) model and design the new antifouling polymer brushes using the group-based supporting vector regression (SVR) model. The resultant two machine learning models not only demonstrated their reliability, predictivity, and applicability, but also established the composition-structure–property relationships using both descriptors and functional groups. Finally, we synthesized different repurposed and newly designed polymer brushes, as predicted by ANN and SVR models, all of which exhibited excellent surface resistance to protein adsorption from undiluted human blood serum and plasma at optimal film thicknesses. Overall, our data-driven machine learning models can be used as an intelligent tool for determining, repurposing, and designing new superior antifouling materials beyond polymer brushes.