Sheikhshoaei Amir Hossein, Khoshsima Ali
School of Petroleum and Chemical Engineering, Hakim Sabzevari University, Sabzevar, Iran.
Center for Atmospheric Research, University of Oulu, 90014, Oulu, Finland.
Sci Rep. 2025 Jul 15;15(1):25465. doi: 10.1038/s41598-025-09600-z.
The multifaceted effects of the presence of thiophenic compounds on the environment are significant and cannot be overlooked. As heterocyclic compounds, thiophene and its derivatives play a significant role in materials science, particularly in the design of organic semiconductors, pharmaceuticals, and advanced polymers. Accurate prediction of their thermophysical properties is critical due to its impact on structural, thermal, and transport properties. This study utilizes state-of-the-art machine learning and deep learning models to predict high-pressure density of seven thiophene derivatives, namely thiophene, 2-methylthiophene, 3-methylthiophene, 2,5-dimethylthiophene, 2-thiophenemethanol, 2-thiophenecarboxaldehyde, and 2-acetylthiophene. The critical properties including critical temperature (T), critical pressure (P), critical volume (V), and acentric factor (ω), together with boiling point (T), and molecular weight (Mw) were used as input parameters. Models employed include Decision Tree (DT), Adaptive Boosting Decision Tree (AdaBoost-DT), Light Gradient Boosting Machine (LightGBM), Gradient Boosting (GBoost), TabNet, and Deep Neural Network (DNN). The statistical error evaluation showed that the LightGBM model showed superior performance with an average absolute percent relative error (AAPRE) of 0.0231, a root mean square error of 0.3499, and coefficient of determination (R) of 0.9999. The leverage method showed that 99.10 percent of the data was valid. These findings highlight the effectiveness of using critical properties as inputs and underscore the potential of the LightGBM model for reliable high-pressure density prediction of thiophene derivatives. This provides a robust tool for advancing materials science applications, and offers valuable insights for material design under extreme conditions.
噻吩类化合物的存在对环境具有多方面的显著影响，不容忽视。作为杂环化合物，噻吩及其衍生物在材料科学中发挥着重要作用，特别是在有机半导体、药物和先进聚合物的设计方面。由于其对结构、热和传输性质的影响，准确预测它们的热物理性质至关重要。本研究利用最先进的机器学习和深度学习模型来预测七种噻吩衍生物的高压密度，这七种衍生物分别是噻吩、2-甲基噻吩、3-甲基噻吩、2,5-二甲基噻吩、2-噻吩甲醇、2-噻吩甲醛和2-乙酰基噻吩。包括临界温度（T）、临界压力（P）、临界体积（V）和偏心因子（ω）在内的临界性质，以及沸点（T）和分子量（Mw）被用作输入参数。所采用的模型包括决策树（DT）、自适应提升决策树（AdaBoost-DT）、轻梯度提升机（LightGBM）、梯度提升（GBoost）、TabNet和深度神经网络（DNN）。统计误差评估表明，LightGBM模型表现出卓越的性能，平均绝对相对百分比误差（AAPRE）为0.0231，均方根误差为0.3499，决定系数（R）为0.9999。杠杆法表明99.10%的数据是有效的。这些发现突出了使用临界性质作为输入的有效性，并强调了LightGBM模型在可靠预测噻吩衍生物高压密度方面的潜力。这为推进材料科学应用提供了一个强大的工具，并为极端条件下的材料设计提供了有价值的见解。
