Universidade de Sao Paulo, ESALQ, Departamento de Ciências Exatas, Piracicaba 13418-900, Brazil.
Universidade de Sao Paulo, ESALQ, Departamento de Ciências Exatas, Piracicaba 13418-900, Brazil; Syngenta Crop Protection AG, Global Biological Data Analytics, 4058 Basel, Switzerland.
Sci Total Environ. 2023 Dec 20;905:167041. doi: 10.1016/j.scitotenv.2023.167041. Epub 2023 Sep 18.
For over a century, ecotoxicological studies have reported the occurrence of hormesis as a significant phenomenon in many areas of science. In plant biology, hormesis research focuses on measuring morphological, physiological, biochemical, and productivity changes in plants exposed to low doses of herbicides. These studies involve multiple features that are often correlated. However, the multivariate aspect and interdependencies among components of a plant system are not considered in the adopted modeling framework. Therefore, a multivariate nonlinear modeling approach for hormesis is proposed, where information regarding correlations among response variables is taken into account through a variance-covariance matrix obtained from univariate residuals. The proposed methodology is evaluated through a Monte Carlo simulation study and an application to experimental data from safflower (Carthamus tinctorius L.) cultivation. In the simulation study, the multivariate model outperformed the univariate models, exhibiting higher precision, lower bias, and greater accuracy in parameter estimation. These results were also confirmed in the analysis of the experimental data. Using the delta method, mean doses of interest can be derived along with their associated standard errors. This is the first study to address hormesis in a multivariate context, allowing for a better understanding of the biphasic dose-response relationships by considering the interrelationships among various measured characteristics in the plant system, leading to more precise parameter estimates.
一个多世纪以来，生态毒理学研究报告称，在许多科学领域，激素作用是一个重要现象。在植物生物学中，激素作用研究侧重于测量暴露于低剂量除草剂的植物的形态、生理、生化和生产力变化。这些研究涉及到许多相互关联的特征。然而，所采用的建模框架并未考虑植物系统各组成部分的多变量方面和相互依存关系。因此，提出了一种用于激素作用的多变量非线性建模方法，该方法通过从单变量残差获得的方差-协方差矩阵来考虑关于响应变量之间相关性的信息。通过蒙特卡罗模拟研究和对红花（Carthamus tinctorius L.）种植的实验数据的应用来评估所提出的方法。在模拟研究中，多变量模型优于单变量模型，在参数估计的精度、偏差和准确性方面表现出更高的精度、更低的偏差和更高的准确性。这些结果在对实验数据的分析中也得到了证实。使用 delta 方法，可以得出感兴趣的平均剂量及其相关标准误差。这是第一项在多变量背景下研究激素作用的研究，通过考虑植物系统中各种测量特征之间的相互关系，可以更好地理解双峰剂量-反应关系，从而得出更精确的参数估计。
