LncRNAs are intermediate molecules that participate in the most diverse biological processes in humans, such as gene expression control and X-chromosome inactivation. Numerous researches have associated lncRNAs with a wide range of diseases, such as breast cancer, leukemia, and many other conditions. In this work, we propose a graph-based method named PANDA. This method treats the prediction of new associations between lncRNAs and diseases as a link prediction problem in a graph. We start by building a heterogeneous graph that contains the known associations between lncRNAs and diseases and additional information such as gene expression levels and symptoms of diseases. We then use a Graph Auto-encoder to learn the representation of the nodes’ features and edges, finally applying a Neural Network to predict potentially interesting novel edges. The experimental results indicate that PANDA achieved a 0.976 AUC-ROC, surpassing state-of-the-art methods for the same problem, showing that PANDA could be a promising approach to generate embeddings to predict potentially novel lncRNA-disease associations.

中文翻译：

---

用于预测新型 lncRNA 疾病关联的图卷积自动编码器

LncRNA 是参与人类最多样化的生物过程的中间分子，例如基因表达控制和 X 染色体失活。许多研究已将 lncRNA 与多种疾病联系起来，例如乳腺癌、白血病和许多其他疾病。在这项工作中，我们提出了一种名为 PANDA 的基于图的方法。该方法将 lncRNA 与疾病之间新关联的预测视为图中的链接预测问题。我们首先构建一个异构图，其中包含 lncRNA 与疾病之间的已知关联以及其他信息，例如基因表达水平和疾病症状。然后，我们使用图形自动编码器来学习节点特征和边缘的表示，最后应用神经网络来预测可能有趣的新边缘。