農藥對鳥類和哺乳動物物種內的急性毒性分析和物種間的急性毒性建模

Abstract

鳥類在體型、生活特徵、飲食偏好、棲息地、行為和代謝酶活性方面表現出比哺乳動物更大的變異性。這些差異可能會影響物種內的毒性變化或導致毒理學數據的空白，通常需要從其他標準物種推斷數據。因此，本研究旨在預測哺乳動物和鳥類的種間毒性數據，並嘗試使用哺乳動物模型跨綱預測鳥類的毒性。 本研究收集了鳥類和哺乳動物共同的19種農藥數據，篩選了影響物種內差異的實驗條件數據，並使用體重與毒性指標LD50之間的非線性關係（異速生長），包括化合物描述符作為變數，建立鳥類和哺乳動物的種間毒性模型。此外，最佳的哺乳動物種間模型被用來嘗試預測鳥類數據。觀察發現，哺乳動物和鳥類的數據在不同化合物中的分佈不均，研究普遍更關注對於綱內毒性較高的化合物。通過異速生長標度，支持鳥類的平均異速生長指數（b）值為1.15，而哺乳動物的平均b值為1。三個模型被用於比較種間毒性預測：W（使用體重作為變數），MD（使用分子描述符作為變數），以及W-MD（結合分子描述符和體重作為變數）。哺乳動物W-MD模型的RMSE為0.41，鳥類W-MD模型的RMSE為0.58，均優於W和MD模型的預測結果。W-MD模型在預測不常見物種的毒性方面也是可行的，哺乳動物模型的RMSE為0.45，鳥類模型的RMSE為0.51，表明模型預測可以減少對不常見動物進行毒性測試的需求。使用哺乳動物W-MD模型預測鳥類毒性的效果較差，RMSE為0.95。這是由於某些相較於鳥類毒性更高的化合物導致預測結果的高估。進一步分析鳥類和哺乳動物毒性差異顯著的化合物將有助於改進跨綱模型。 本研究的應用不僅限於農藥，還包括其他化學品。通過提高毒性預測的準確性，本研究可在環境管理、公共衛生和藥理學方面受益，為政策制定者和風險評估者提供關於農藥和化學品的監管、野生鳥類的保護等方面的有力工具。

Birds exhibit greater variability in body size, life characteristics, dietary preferences, habitats, behaviors, and metabolic enzyme activity compared to mammals. These differences may influence variations in toxicity within species or result in gaps in toxicological data, often necessitating data extrapolation from other standard species. Therefore, this study aims to predict interspecies toxicity data for mammals and birds and to attempt cross-class predictions of avian toxicity using mammalian models. This study collects data on 19 pesticides common to both birds and mammals, filters experimental condition data that affect within-species differences, and uses the nonlinear relationship (allometric) between body weight and the toxicity indicator LD50, including compound descriptors as variables, to establish interspecies toxicity models for birds and mammals. Furthermore, the best interspecies model for mammals is used to attempt predictions of avian data. It was observed that data for mammals and birds are unevenly distributed across different compounds, with research generally focusing more on compounds with higher toxicity within the class. Through allometric scaling, it is also supported that the average allometric exponent (b) value for birds is 1.15, while the average b value for mammals is 1. Three models were compared for interspecies: W (using body weight as variable), MD (using molecular descriptors as variables), and W-MD (combining molecular descriptors and body weight as variables). The RMSE of the mammalian W-MD model is 0.41, and the avian W-MD model is 0.58, both of which are better than the W and MD model predictions. The W-MD models are also feasible for predicting toxicity in uncommon species. The mammalian model has an RMSE of 0.45, and the avian model has an RMSE of 0.51, indicating that the model predictions can reduce the need for toxicity testing in uncommon animals. The performance of predicting avian toxicity by mammalian W-MD model to is poor, with an RMSE of 0.95. This is due to certain compounds that are more toxic to birds, causing overestimation in the predictions. Further analysis of compounds with significant toxicity differences between birds and mammals would improve the cross-class model. This study's applications are not limited to pesticides but also other chemicals. It can benefit environmental management, public health, and pharmacology by improving the accuracy of toxicity predictions and offering robust tools for policymakers and risk assessors regarding regulatory of pesticides and chemicals, conservation of wild birds, and the protection