IAA-NCEA/EPA-0001; “Chemoinformatics of EPA’s EDSP Universe of Chemicals”

为IAA/USEPA的两个DNTP项目提供了化学信息支持;第一个项目使用EPA的ECOTOX数据库，第二个项目使用EPA的NCEA（国家中心环境评估）。 (1)生态毒理学知识库（ECOTOX）是一个全面的、可公开获取的知识库，提供关于水生生物、陆生植物和野生动物的单一化学品环境毒性数据。为了简化这一劳动密集型过程，Sciome评估了使用机器学习方法根据相关性自动对文档进行分类的可行性，并确定排除那些参考文献的排除理由。在过去30年的数据库增长和发展过程中，EPA在ECOTOX数据库中手动管理了约45，000篇参考文献。目前的数据管理流程包括几个手动和耗时的步骤。通过使用人工智能（人工智能）软件工具，该过程可以大大缩短并变得更加一致，并且启动了一个开发ECOTOX词典和开发自动化策展策略的过程。 (2)在与NCEA合作的第二个项目中，对EDSP化学品宇宙进行了分析，以形成一个数据库门户。 最初的聚类练习是在大约700种化合物上完成的，这些化合物是完全明确的。 进一步的聚类扩展到化合物（~ 14 K）的联合，通过在CompTox仪表板中搜索EPA共享文件中提供的CAS编号和名称获得。每个聚类实验涉及完整的数据清洗和协调，结构量化的几个指纹和描述符，然后调查几个距离测量和聚类方法。 一些类特定的聚类实验的方法，以确定简约集的集群进行了追求。 在权衡了开发EDSP门户的几个可用平台的利弊后，EPA和Sciome同意在Python/Django平台上开发EDSP门户。该门户网站最初将存放一些物理化学数据和一级生物测定数据的化合物，迄今已明确策划。门户网站的功能可以通过定期向客户展示和反馈的周期不断得到加强。

Chemoinformatic support for two IAA/USEPA projects with DNTP were performed; first project with EPA's ECOTOX database and the second project with EPA's NCEA (Natl Ctr Environmental Assessment). (1) The ECOTOXicology Knowledgebase (ECOTOX) is a comprehensive, publicly available knowledgebase providing single chemical environmental toxicity data on aquatic life, terrestrial plants and wildlife. To streamline this labor-intensive process, Sciome evaluated the feasibility of using machine learning methods to automatically classify documents according to relevance, and to identify the exclusion rationale for those references that are excluded. There are ~45,000 references that have been manually curated by EPA in ECOTOX database over the course of the last thirty years of database growth and development. The current data curation process consists of several manual and time-consuming steps. The process could be greatly shortend and made more consistent by use of AI (artificial intelligence) software tools and a process was initiated to develop an ECOTOX lexicon and develop a strategy for automated curation. (2) For the 2nd project with NCEA, the EDSP Universe of Chemicals was analyzed towards forming a database portal. An initial clustering exercise was completed on ~700 compounds that were fully and unambiguously curated. Further clustering was extended to the union of compounds (~14K) obtained by searching the CompTox Dashboard for CAS numbers and Names provided in the file shared by EPA. Each clustering experiment involved complete data cleaning and harmonization, structure quantification by several fingerprints and descriptors followed by investigating several distance measurements and clustering methods. Some class-specific clustering experiments with methods to identify parsimonious set of clusters were pursued. After weighing pros and cons of several available platforms for developing the EDSP portal, EPA and Sciome agreed that the EDSP Portal would be developed on the Python/Django platform. The Portal would initially house some physical-chemistry data and Tier-1 bio-assay data for the compounds that have so far been unambiguously curated. The functionality of the Portal could be continually enhanced through a regular show-and-feedback cycle with the clients.