ToxIndex-CPG: Machine learning driven platform integrating a hazard susceptibility database to quantify chemical toxicity factors, predict risk levels and classify biological responses

2019年全球毒理学测试整体市场为81亿美元，预计将达到270亿美元 到2025年由于毒理学测试是大多数产品开发的先决条件，因此会增加大量时间， 在没有获得关键数据的情况下，这些数据还代表人类健康危害。为了最大限度地减少时间， 市场，费用和动物使用，计算生物学和机器学习（ML）的进步正在帮助 进行更有效的计算机模拟。这些战略正在推动先进的 计算工具。更具体地说，目前在6350亿美元的CPG市场中存在强大的价值主张 用于确保安全并通过将毒理学危害简介与 生殖健康与化学品、接触和产品使用案例的关系。这将有助于更好地理解和 将化学品优先纳入产品，以尽量减少相关的生殖健康危害。 ToxIndex-CPG平台将通过基于Web的界面解决这一不断增长的市场需求， CPG毒理学研究人员可以访问定制数据，用于早期产品规划和研究设计。平台 将专注于数据库的持续管理，以维护现有文献和数据中的已知关系 源，以及用于未知组合的预测关系的高级算法。该项目将 针对CPG产品和生殖健康危害，因为这是弱势群体的主要市场和风险 人口。用户前端将被设计为一个基于Web的工具，供毒理学研究人员查询特定的 化学品、CPG用例和健康危害。根据查询输入，平台将返回一个排序和排名的 潜在的不良生殖健康后果清单。研究人员将能够探索特定的 通过先进的可视化工具对化学品的生殖危害进行分级。风险关系 化学品和人为因素以及计划的CPG产品用例将通过ML使用定量 构效关系（QSAR）模型。该平台将利用现有的数据源， 医疗数据来构建模型，并随着数据集的不断增长而不断自适应学习。该平台将 优先考虑应用程序编程接口（API），以支持不断增长的化学信息学开发人员市场。 第一阶段将针对数据聚合、ML开发和原型接口的可行性。发展将 利用现有工具Sysrev从出版物和数据源中自动提取数据， 成功的可能性。Sysrev平台将解析现有数据源，以提取已知的人为因素， 给定化学毒物和生殖健康危害的用例敏感性因素。这将创建一个 已知因素的初始危害数据库作为ML测试的金标准。接下来，QSAR ML模型将 开发了将化学品与危害相关联的调解模型，然后从化学品到危害， 了解特定人为因素和这些化学品的使用案例中的因果关系可能性。最后，一个原型 网络应用程序和可视化将在毒理学市场用户的可用性研究中开发和部署。

The overall global market for toxicology testing was $8.1 billion in 2019 and is expected to reach $27 billion by 2025. As toxicological testing is a pre-requisite step in most product development, it adds significant time and costs, as well as represents human health hazards when key data is not captured. In order to minimize time to market, expense, and animal use, advances in computational biology and machine learning (ML) are helping conduct more efficient in-silico simulations. These strategies are driving strong growth for advanced computational tools. More specifically, there is currently a strong value proposition in the $635 billon CPG market for tools that ensure safety & expedite product design strategies by linking toxicology hazard profiles in reproductive health to chemicals, exposure & product use cases. This will allow a better understanding and prioritization of chemicals for integration in products to minimize associated reproductive health hazards. The ToxIndex-CPG platform will solve this growing market need through a web-based interface that allows CPG toxicology researchers access to customized data for early product planning & study design. The platform will focus on continuous curation of a database to maintain known relationships in existing literature and data sources, as well as advanced algorithms for predictive relationships for unknown combinations. This project will target CPG products and reproductive health hazards, as this represents major markets & risks to vulnerable populations. The user front end will be designed as a web-based tool for toxicology researchers to query specific chemicals, CPG use cases, & health hazards. Based on query inputs, the platform will return a sorted and ranked list of potential adverse reproductive health outcomes. Researchers will be able to explore impact of specific chemicals on ranked reproductive hazards through advanced visualization tools. Hazard relationships between chemicals and human factors & planned CPG product use cases will be learned through ML using quantitative structure-activity relationship (QSAR) models. The platform will leverage existing data sources for chemical & medical data to build models & continue to adaptively learn as datasets continue to grow. The platform will prioritize application programming interfaces (API) to support a growing market of cheminformatics developers. Phase I will target feasibility of data aggregation, ML development, & prototype interface. Development will leverage an existing tool, Sysrev, for automated data extraction from publications & data sources to increase likelihood of success. The Sysrev platform will parse existing data sources to extract known human factors and use case susceptibility factors for a given chemical toxicant and reproductive health hazards. This will create an initial hazard database of known factors as a gold standard for ML testing. Next, QSAR ML models will be developed to associate chemicals to hazards, and then mediation models from chemicals through hazards to understand causality likelihood in specific human factors and use cases of those chemicals. Finally, a prototype web app and visualizations will be developed and deployed in a usability study with toxicology market users.