摘要:一直以来,从新药物的发现到最终应用的过程被认为是非常耗时且消耗资源密集的。在化学领域,经典传统方法密度泛兩理论(DFT)使用非常广泛,其计算出分子的密度泛兩并推导出各种性质。然面,传统的量子模拟技术既昂贵又难以探索潜在大范围的掺杂分子结构。为了降低成本并提高效率,提出了一种基于化学环境的图神经网络权型,希望能在新型材料和药物的研发上推动发展。本文探索领域聚焦于石墨怎化碳(g-C3N4)及其掺杂变体。鉴于石墨氨化碳(E-C3N4)的分子性质带院在现实中的重要性,准确預测材料的光学帶隙成为了本研究的目标。本文使用基于化学环境的图神经网络有效地捕捉了分子的复杂结构,即使同时探索具有多个变体的掺杂:-C3N4结构,它也能精确预测它们的带隙,相比于传统的图神经网络有极大的提升.提供了一种方便快捷且精确的工具。

Abstraet: The process from the discovery of new drugs to their final application has always been consideredvery time-consuming and resource intensive. In the field of chemistry, the classical traditional method densityfunctional theory (DFT) is widely used, which calculates the density functional of molecules and derives variousproperties. However, traditional quantum simulation techniques are both expensive and difficult to explorepotential large-seale doped molecular structures. In order to reduce costs and improve eficiency, this articleproposes a graph neural network model based on chemical environment, hoping to promote development in theresearch and development of new materials and drugs. The field explored in this article focuses on graphite nitridccarbon (g-C3N4) and its doped variants. Given the importance of the molecular properties and bandgap ofgraphite nitride carbon (g-C3N4) in reality, accurately predicting the optical bandgap of the material has becomcthe research objective of this paper. This article effectively captures the complex structure of molecules usinga graph neural network based on chemical environment. Even when exploring doped g-C3N4 structures withmultiple variants simultaneously, it can accurately predict their band gaps, which is greatly improved compared to traditional graph neural networks, providing a convenient, fast, and accurate tool.