Parvez Mohammad Anwar, Mehedi Ibrahim M
Department of Chemical Engineering, College of Engineering, King Faisal University, Al-Ahsa 31982, Saudi Arabia.
School of Robotics, XJTLU Entrepreneur College (Taicang), Xi'an Jiaotong-Liverpool University, No. 111 Taicang Ave., Taicang, Suzhou 215400, China.
Polymers (Basel). 2025 Jun 28;17(13):1801. doi: 10.3390/polym17131801.
Polymers have a wide range of applications in materials science, chemistry, and biomedical domains. Conventional design methods for polymers are mostly event-oriented, directed by intuition, experience, and abstract insights. Nevertheless, they have been effectively utilized to determine several essential materials; these techniques are facing important challenges owing to the great requirement of original materials and the huge design area of organic polymers and molecules. Enhanced and inverse materials design is the best solution to these challenges. With developments in high-performing calculations, artificial intelligence (AI) (particularly Deep learning (DL) and Machine learning (ML))-aided materials design is developing as a promising tool to show development in various domains of materials science and engineering. Several ML and DL methods are established to perform well for polymer classification and detection presently. In this paper, we design and develop a Simplified Molecular Input Line Entry System Based Polymer Property Detection and Classification Using Pareto Optimization Algorithm (SMILES-PPDCPOA) model. This study presents a novel deep learning framework tailored for polymer property classification using SMILES input. By integrating a one-dimensional convolutional neural network (1DCNN) with a gated recurrent unit (GRU) and optimizing the model via Pareto Optimization, the SMILES-PPDCPOA model demonstrates superior classification accuracy and generalization. Unlike existing methods, our model is designed to capture both local substructures and long-range chemical dependencies, offering a scalable and domain-specific solution for polymer informatics. Furthermore, the proposed SMILES-PPDCPOA model executes a one-dimensional convolutional neural network and gated recurrent unit (1DCNN-GRU) technique for the classification process. Finally, the Pareto optimization algorithm (POA) adjusts the hyperparameter values of the 1DCNN-GRU algorithm optimally and results in greater classification performance. Results on a benchmark dataset show that SMILES-PPDCPOA achieves an average classification accuracy of 98.66% (70% Training, 30% Testing) across eight polymer property classes, with high precision and recall metrics. Additionally, it demonstrates superior computational efficiency, completing tasks in 4.97 s, outperforming other established methods such as GCN-LR and ECFP-NN. The experimental validation highlights the potential of SMILES-PPDCPOA in polymer property classification, making it a promising approach for materials science and engineering. The simulation result highlighted the improvement of the SMILES-PPDCPOA system when compared to other existing techniques.
聚合物在材料科学、化学和生物医学领域有着广泛的应用。传统的聚合物设计方法大多以事件为导向，由直觉、经验和抽象见解指导。然而，它们已被有效地用于确定几种重要材料；由于对原材料的巨大需求以及有机聚合物和分子的巨大设计领域，这些技术正面临着重大挑战。增强和逆向材料设计是应对这些挑战的最佳解决方案。随着高性能计算的发展，人工智能（AI）（特别是深度学习（DL）和机器学习（ML））辅助的材料设计正在发展成为一种有前途的工具，以展示材料科学与工程各个领域的发展。目前已经建立了几种ML和DL方法，在聚合物分类和检测方面表现良好。在本文中，我们设计并开发了一种基于简化分子输入线性输入系统，使用帕累托优化算法的聚合物性能检测与分类（SMILES-PPDCPOA）模型。本研究提出了一种新颖的深度学习框架，用于使用SMILES输入进行聚合物性能分类。通过将一维卷积神经网络（1DCNN）与门控循环单元（GRU）集成，并通过帕累托优化对模型进行优化，SMILES-PPDCPOA模型展示了卓越的分类准确率和泛化能力。与现有方法不同，我们的模型旨在捕捉局部子结构和长程化学依赖性，为聚合物信息学提供了一种可扩展的、特定领域的解决方案。此外，所提出的SMILES-PPDCPOA模型在分类过程中执行一维卷积神经网络和门控循环单元（1DCNN-GRU）技术。最后，帕累托优化算法（POA）对1DCNN-GRU算法的超参数值进行了最优调整，从而获得了更高的分类性能。在一个基准数据集上的结果表明，SMILES-PPDCPOA在八个聚合物性能类别上实现了平均分类准确率为98.66%（70%训练，30%测试），具有高精度和召回率指标。此外，它还展示了卓越的计算效率，在4.97秒内完成任务，优于其他已建立的方法，如GCN-LR和ECFP-NN。实验验证突出了SMILES-PPDCPOA在聚合物性能分类方面的潜力，使其成为材料科学与工程中一种有前途的方法。模拟结果突出了SMILES-PPDCPOA系统与其他现有技术相比的改进。
