Bren School of Environmental Science and Management, University of California , Santa Barbara, California 93106, United States.
Environ Sci Technol. 2017 Sep 19;51(18):10777-10785. doi: 10.1021/acs.est.7b02862. Epub 2017 Aug 30.
The number of chemicals in the market is rapidly increasing, while our understanding of the life-cycle impacts of these chemicals lags considerably. To address this, we developed deep artificial neural network (ANN) models to estimate life-cycle impacts of chemicals. Using molecular structure information, we trained multilayer ANNs for life-cycle impacts of chemicals using six impact categories, including cumulative energy demand, global warming (IPCC 2007), acidification (TRACI), human health (Impact2000+), ecosystem quality (Impact2000+), and eco-indicator 99 (I,I, total). The application domain (AD) of the model was estimated for each impact category within which the model exhibits higher reliability. We also tested three approaches for selecting molecular descriptors and identified the principal component analysis (PCA) as the best approach. The predictions for acidification, human health, and the eco-indicator 99 model showed relatively higher performance with R values of 0.73, 0.71, and 0.87, respectively, while the global warming model had a lower R of 0.48. This study indicates that ANN models can serve as an initial screening tool for estimating life-cycle impacts of chemicals for certain impact categories in the absence of more reliable information. Our analysis also highlights the importance of understanding ADs for interpreting the ANN results.
市场上的化学物质数量正在迅速增加，而我们对这些化学物质生命周期影响的理解却大大滞后。为了解决这个问题，我们开发了深度人工神经网络 (ANN) 模型来估计化学物质的生命周期影响。我们使用分子结构信息，针对包括累积能源需求、全球变暖 (IPCC 2007)、酸化 (TRACI)、人类健康 (Impact2000+)、生态系统质量 (Impact2000+) 和生态指标 99 (I,I, 总) 在内的六个影响类别，训练了用于化学物质生命周期影响的多层 ANN。模型的应用领域 (AD) 是针对每个影响类别估计的，在这些类别中，模型表现出更高的可靠性。我们还测试了三种选择分子描述符的方法，并确定主成分分析 (PCA) 是最佳方法。对于酸化、人类健康和生态指标 99 模型，预测表现出相对较高的性能，R 值分别为 0.73、0.71 和 0.87，而全球变暖模型的 R 值较低，为 0.48。本研究表明，在缺乏更可靠信息的情况下，ANN 模型可以作为估计某些影响类别化学物质生命周期影响的初始筛选工具。我们的分析还强调了理解 AD 对于解释 ANN 结果的重要性。
