Liu Yixin, Jiang Guowei, Sun Miaomiao, Zhou Ziyan, Liang Pengchen, Chang Qing
School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai, 200093, China.
Pharmacy Department, Jiading District Central Hospital Affiliated Shanghai University of Medicine and Health Sciences, Shanghai, 201318, China.
Curr Comput Aided Drug Des. 2025;21(1):28-37. doi: 10.2174/0115734099266731231115065030.
Osteoporosis (OP) is one of the most common diseases in the elderly population. It is mostly treated with medication, but drug research and development have the disadvantage of taking a long time and having a high cost.
Therefore, we developed a graph neural network with the help of artificial intelligence to provide new ideas for drug research and development for OP.
In this study, we built a new osteoporosis graph (called OPGraph) and proposed a deep graph neural network (called DeepTransformer) to predict new drugs for OP. OPGraph is a graph data model established by gathering features and their interrelationships from a vast amount of OP data. DeepTransformer uses GraphTransformer as its foundational network and applies residual connections for deep layering.
The analysis and results showed that DeepTransformer outperformed numerous models on OPGraph, with area under the curve (AUC) and area under the precision-recall curve (AUPR) reaching 0.9916 and 0.9911, respectively. In addition, we conducted an validation experiment on two of the seven predicted compounds (Puerarin and Aucubin), and the results corroborated the predictions of our model.
The model we developed with the help of artificial intelligence can effectively reduce the time and cost of OP drug development and reduce the heavy economic burden brought to patient's family by complications caused by osteoporosis.
骨质疏松症（OP）是老年人群中最常见的疾病之一。其治疗大多采用药物，但药物研发存在耗时久、成本高的缺点。
因此，我们借助人工智能开发了一种图神经网络，为OP的药物研发提供新思路。
在本研究中，我们构建了一个新的骨质疏松症图（称为OPGraph），并提出了一种深度图神经网络（称为DeepTransformer）来预测治疗OP的新药。OPGraph是通过从大量OP数据中收集特征及其相互关系而建立的图数据模型。DeepTransformer以GraphTransformer作为其基础网络，并应用残差连接进行深度分层。
分析与结果表明，DeepTransformer在OPGraph上的表现优于众多模型，曲线下面积（AUC）和精确率-召回率曲线下面积（AUPR）分别达到0.9916和0.9911。此外，我们对七种预测化合物中的两种（葛根素和桃叶珊瑚苷）进行了验证实验，结果证实了我们模型的预测。
我们借助人工智能开发的模型可以有效减少OP药物研发的时间和成本，并减轻骨质疏松症并发症给患者家庭带来的沉重经济负担。
