TransPolymer: a Transformer-based language model for polymer property predictions

作     者：Changwen Xu Yuyang Wang Amir Barati Farimani 

作者机构：Department of Materials Science and EngineeringCarnegie Mellon University5000 Forbes AvePittsburgh 15213 PAUSA Department of Mechanical EngineeringCarnegie Mellon University5000 Forbes AvePittsburgh 15213 PAUSA Machine Learning DepartmentCarnegie Mellon University5000 Forbes AvePittsburgh 15213 PAUSA Department of Chemical EngineeringCarnegie Mellon University5000 Forbes AvePittsburgh 15213 PAUSA 

出 版 物：《npj Computational Materials》 (计算材料学（英文）)

年 卷 期：2023年第9卷第1期

页      面：1703-1716页

核心收录：

学科分类：081704[工学-应用化学] 07[理学] 08[工学] 0817[工学-化学工程与技术] 081203[工学-计算机应用技术] 070305[理学-高分子化学与物理] 080501[工学-材料物理与化学] 0805[工学-材料科学与工程（可授工学、理学学位）] 0703[理学-化学] 0835[工学-软件工程] 0701[理学-数学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：U.S. Department of Energy, USDOE, (DE-AR0001221) Advanced Research Projects Agency - Energy, ARPA-E Carnegie Mellon University, CMU 

主　　题：property polymer rational 

摘      要：Accurate and efficient prediction of polymer properties is of great significance in polymer ***,expensive and time-consuming experiments or simulations are required to evaluate polymer ***,Transformer models,equipped with self-attention mechanisms,have exhibited superior performance in natural language ***,such methods have not been investigated in polymer ***,we report TransPolymer,a Transformer-based language model for polymer property *** proposed polymer tokenizer with chemical awareness enables learning representations from polymer *** experiments on ten polymer property prediction benchmarks demonstrate the superior performance of ***,we show that TransPolymer benefits from pretraining on large unlabeled dataset via Masked Language *** results further manifest the important role of self-attention in modeling polymer *** highlight this model as a promising computational tool for promoting rational polymer design and understanding structure-property relationships from a data science view.