Elsenety Mohamed M
Department of Chemistry, Faculty of Science, Al-Azhar University, Nasr City, Cairo, 11884, Egypt.
Sci Rep. 2024 Nov 3;14(1):26539. doi: 10.1038/s41598-024-76739-6.
This scientific paper presents a novel approach to explore and predict the potential of imidazole-based organic dyes for use in Dye-Sensitized Solar Cells (DSSCs) using a machine learning web application. The design of efficient and cost-effective organic dyes is critical to enhance the performance of DSSCs. Traditional experimental methods are time-consuming and resource-intensive, making it challenging to screen a large number of potential dyes. In this study, we propose a machine learning-based approach to accelerate the discovery process by predicting the photovoltaic performance of imidazole-based organic dyes. Machin learning predictions provide valuable insights into the expected PCE% and behaviors of the molecules toward DSSCs. Based on the RDKit library, several fingerprints such as Molecular ACCess System, Avalon, Daylight, Pharmacophore and Morgan with different radius (r2, r3, r4), were studied. In addition, more than 20 ML algorithms using different cross validation (3, 5, 7, 10) were also evaluated. Among of these, Deep Neural Network models of MLPRegressor algorithm based on the daylight fingerprint shows a significant coefficient of determination combined with the lowest errors. Utilize the trained ML models to screen of 50 million SMILE structure for identify promising imidazole and nitrogen-containing derivative as a doner group. By replacing the donor groups in the well-known MK2 dye structure with the top imidazole derivatives proposed by machine learning, significant improvements in PCE were observed, increasing from 7.70% to as high as 11.49%, representing nearly a 50% enhancement over the control. DFT calculations confirm the ML predictions and clarify the significantly higher oscillator strength and charge transfer properties of MK2-DM1, compared to MK2. This result provides a promising pathway for developing new dye materials that can push the efficiency limits of DSSCs, leading to more efficient solar energy conversion technologies in the future. In addition, a developed web application offers a user-friendly interface for researchers to input their molecular structures and obtain PCE% predictions toward DSSCs. This information can guide researchers in designing a new imidazole dye with high photovoltaic performance to validate and refine the predictions without time consuming.
这篇科学论文提出了一种新颖的方法，利用机器学习网络应用程序来探索和预测基于咪唑的有机染料在染料敏化太阳能电池（DSSC）中的应用潜力。设计高效且经济高效的有机染料对于提高DSSC的性能至关重要。传统的实验方法既耗时又耗费资源，因此筛选大量潜在染料具有挑战性。在本研究中，我们提出了一种基于机器学习的方法，通过预测基于咪唑的有机染料的光伏性能来加速发现过程。机器学习预测为预期的光电转换效率百分比（PCE%）以及分子在DSSC中的行为提供了有价值的见解。基于RDKit库，研究了几种指纹，如分子访问系统、阿瓦隆、日光、药效团和不同半径（r2、r3、r4）的摩根指纹。此外，还评估了使用不同交叉验证（3、5、7、10）的20多种机器学习算法。其中，基于日光指纹的MLPRegressor算法的深度神经网络模型显示出显著的决定系数，且误差最低。利用训练好的机器学习模型筛选5000万个SMILE结构，以识别有前景的咪唑和含氮衍生物作为供体基团。通过用机器学习提出的顶级咪唑衍生物取代著名的MK2染料结构中的供体基团，观察到PCE有显著提高，从7.70%提高到高达11.49%，比对照提高了近50%。密度泛函理论（DFT）计算证实了机器学习的预测，并阐明了与MK-2相比，MK2-DM1具有显著更高的振子强度和电荷转移特性。这一结果为开发能够突破DSSC效率极限的新型染料材料提供了一条有前景的途径，从而在未来带来更高效的太阳能转换技术。此外，开发的网络应用程序为研究人员提供了一个用户友好的界面，用于输入他们的分子结构并获得对DSSC的PCE%预测。这些信息可以指导研究人员设计具有高光伏性能的新型咪唑染料，以便在不耗费时间情况下验证和完善预测。
