High-throughput screening of chemicals as functional substitutes using structure-based classification models
High-throughput screening of chemicals as functional substitutes using structure-based classification models
复制标题
DOI：
10.1039/c6gc02744j
复制
发表时间：
2017-02-21
期刊：
影响因子：
9.8
通讯作者：
Isaacs, Kristin K.
中科院分区：
文献类型：
--
作者：
Phillips, Katherine A.;Wambaugh, John F.;Isaacs, Kristin K.
确定在产品中提供特定功能的化学品，但对人体或环境的影响最小，是大多数配方化学家和工程师实践绿色化学的目标。我们提出了一种方法来确定潜在的化学功能的替代品，从大型化学品库使用机器学习为基础的模型。我们收集和分析有关消费品或工业过程中化学品功能的公开信息，以确定一套适用于建模的协调功能类别。我们使用这些化学品的结构和物理化学描述符，建立41个定量结构-使用关系（QSUR）模型，使用随机森林分类的协调功能类别。我们应用这些模型筛选了近6400种具有潜在功能替代品结构信息的化学品库。使用我们的功能用途数据库（Fuse），我们可以确定3121种化学品的用途; 4412种预测功能用途的概率为80%或更高。我们证明了潜在的应用模型，高通量（HT）筛选“候选替代品”合并有效的功能替代分类与危害指标开发HT筛选试验的生物活性。一个描述符集可以获得6356毒素21化学品，已经经历了电池的HT在体外生物活性筛选测定。通过应用QSUR，我们能够确定超过1600种候选化学替代品。这些QSUR可快速应用于数千种其他化学品，以生成HT功能用途信息，并与补充HT毒性信息相结合，用于筛选更环保的化学品替代品。
Identifying chemicals that provide a specific function within a product, yet have minimal impact on the human body or environment, is the goal of most formulation chemists and engineers practicing green chemistry. We present a methodology to identify potential chemical functional substitutes from large libraries of chemicals using machine learning based models. We collect and analyze publicly available information on the function of chemicals in consumer products or industrial processes to identify a suite of harmonized function categories suitable for modeling. We use structural and physicochemical descriptors for these chemicals to build 41 quantitative structure-use relationship (QSUR) models for harmonized function categories using random forest classification. We apply these models to screen a library of nearly 6400 chemicals with available structure information for potential functional substitutes. Using our Functional Use database (FUse), we could identify uses for 3121 chemicals; 4412 predicted functional uses had a probability of 80% or greater. We demonstrate the potential application of the models to high-throughput (HT) screening for "candidate alternatives" by merging the valid functional substitute classifications with hazard metrics developed from HT screening assays for bioactivity. A descriptor set could be obtained for 6356 Tox21 chemicals that have undergone a battery of HT in vitro bioactivity screening assays. By applying QSURs, we were able to identify over 1600 candidate chemical alternatives. These QSURs can be rapidly applied to thousands of additional chemicals to generate HT functional use information for combination with complementary HT toxicity information for screening for greener chemical alternatives.