In this paper, A dual-mode network(CAT-MPNN), which could extract features from molecular graphs and textual information simultaneously, is developed to predict the density of Deep Eutectic Solvents (DESs) for the first time. Deep learnings have been applied to predict the density of DESs. However, a comprehensive collection of large-volume datasets and a dual-mode network capable of processing both molecular and textual information as inputs are still lacking. Thus, a database containing 4440 density experimental data of 726 DESs is established in this work. Meanwhile, a dual-mode network CAT-MPNN based on the Keras framework is been developed to input molecular graph and textual information. Finally, the constructed database is employed to validate the CAT-MPNN which has a R² value of 0.9504, a good robustness and generalizability. Both the database and the source code are available at https://github.com/SereinQAQ/CAT-MPNN.

中文翻译：

---

CAT-MPNN：一种双模式网络，用于预测形成开放访问数据库的深共晶溶剂的密度


本文首次开发了一种可以同时从分子图和文本信息中提取特征的双模网络 （CAT-MPNN） 来预测深共晶溶剂 （DES） 的密度。深度学习已被应用于预测 DES 的密度。然而，仍然缺乏能够同时处理分子和文本信息作为输入的大容量数据集和双模式网络。因此，本研究建立了一个包含 726 个 DES 的 4440 个密度实验数据的数据库。同时，开发了一种基于 Keras 框架的双模网络 CAT-MPNN 来输入分子图和文本信息。最后，利用构建的数据库验证 CAT-MPNN，其 R² 值为 0.9504，具有良好的鲁棒性和泛化性。数据库和源代码都可以在 https://github.com/SereinQAQ/CAT-MPNN 获得。