Analysis of Quantitative High Throughput Screening Data

要更好地了解一种物质诱发毒理学反应的能力，最好是分析在广泛浓度（或剂量）范围内的反应概况，而不是评估在单一浓度（或剂量）下发生的效应。体外qHTS试验是多浓度实验，在国家毒理学计划的努力中发挥重要作用，将毒理学从主要是疾病特异性模型水平的观察科学推进到主要是基于广泛包括靶点特异性、基于机制的生物学观察的预测科学。对qHTS数据的分析在很大程度上是出于对制药应用的保守关注（即尽量减少第一类错误的风险），并且通常依赖于启发式而不是统计测试来进行活动调用。为了评估qHTS研究中的活性，我们开发了一个两阶段决策树统计模型，并将其应用于12种基于细胞的激动剂核受体测定（AR、ER、FXR、GR、LXR、PPARd、PPARg、RXR、TRb、VDR、pxr -人、pxr -大鼠）和p53和ELG1激动剂测定的标准化浓度反应数据。我们还将该模型应用于各种鸡细胞系的核受体拮抗剂（AR、FXR、GR、LXR、PPARd、PPARg、RXR、TRb、VDR）和细胞毒性试验。在第一阶段，从1408种物质中获得的数据拟合到四参数Hill方程和总体f检验，比较Hill方程的最佳拟合和水平线（无反应），每种物质表现出至少25%的功效，使用不同的显著性阈值，在测试浓度范围（5 × 10-10 M至10-4 M）内确定活性化合物。在第二阶段，将第一阶段未检测到活性的化合物与控制值的分布进行比较，alpha值为0.05。使用这种方法，我们确定了更多的活性化合物在每个分析比以前使用的启发式方法。

The ability of a substance to induce a toxicological response is better understood by analyzing the response profile over a broad range of concentrations (or doses) rather than by evaluating effects that occur at a single concentration (or dose). In vitro qHTS assays are multiple-concentration experiments that play an important role in NTPs efforts to advance toxicology from a predominantly observational science at the level of disease-specific models to a predominantly predictive science based on broad inclusion of target-specific, mechanism-based, biological observations. The analysis of qHTS data has largely been motivated by the conservative focus of pharmaceutical applications (i.e., minimizing the risk of Type I error) and generally has relied on heuristics rather than statistical tests to make activity calls. To evaluate the activity within qHTS studies, we developed a two-stage decision tree statistical model and applied it to normalized concentration-response data from twelve cell-based agonist nuclear receptor assays (AR, ER, FXR, GR, LXR, PPARd, PPARg, RXR, TRb, VDR, PXR-human, PXR-rat), and agonist assays for p53 and ELG1. We also applied the model to nuclear receptor antagonist assays (AR, FXR, GR, LXR, PPARd, PPARg, RXR, TRb, VDR) and cytotoxicity assays from various chicken cell lines. In the first stage, data obtained from 1408 substances were fit to a four-parameter Hill equation and an overall F-test comparing the best fit to the Hill equation and a horizontal line (no response) for each substance exhibiting at least 25% efficacy, using different significance thresholds, identified active compounds within the tested concentration range (5 x 10-10 M to 10-4 M). In the second stage, compounds not detected as active in the first stage were evaluated by comparing the distribution of measured responses to a control value, at an alpha value of 0.05. Using this approach, we identified a greater number of active compounds in each assay than a previously utilized heuristic approach.