PatCID: an open-access database of chemical structures in patent documents

PatCID is a chemical-structure database automatically created from images in patent documents. It contains 13M unique molecules, 80M molecule images, and 1.2M annotated documents from the United States (USPTO), Europe (EPO), Japan (JPO), Korea (KIPO), and China (CNIPA). Leveraging state-of-the-art document understanding models, PatCID enables accurate document and molecule retrieval in patents. Examples of how to use PatCID can be found on the PatCID GitHub repositoryPatCID是一个从专利文献图像中自动构建的化学结构数据库（chemical-structure database）。该库包含1300万种独特分子、8000万张分子图像，以及来自美国（USPTO）、欧洲（EPO）、日本（JPO）、韩国（KIPO）和中国（CNIPA）的120万份标注文档。依托当前最先进的文档理解模型，PatCID可实现专利文档与分子的精准检索。PatCID的使用示例可在其GitHub仓库中查阅。