Using surrogate passive sampler devices and predictive machine learning algorithms to replace invertebrate use in micropollutant bioconcentration test

被动采样装置（psd）通常用于监测环境中微污染物的长期发生，主要是疏水性有机化学品（hoc）。psd通常由聚合吸附剂组成，并通过被动扩散收集溶质。最近，psd已被用作鱼类/无脊椎动物在有机碳生物浓度研究中的潜在替代品（logKow 4-6）1。最近，psd被专门用于极性有机化学品（POCs）。例如，药品和个人护理产品（PPCPs）的logKow为-1至4，并且具有多种电离状态。这使得ppcp的建模更具挑战性。对7000种药物（不包括其代谢物和转化产物）进行基于实验室的风险评估是不切实际的，而且成本很高。在这个项目中有一个令人兴奋的机会，可以开发和应用预测方法来优先考虑实验室测试或完全消除使用模式生物的需要。2016年，我们首次成功地在PSDs1上建模并预测了POC吸收速率常数（Rs）。我们最近还使用机器学习本身来预测G. pulex中PPCP的生物浓度，并在之前BBSRC资助的CASE学生（BB/K501177/1）中取得了一些有限的成功。因此，现在是时候将这些知识扩展到3r类型的预测方法，以用于现在结合psd的额外POC生物浓度研究，以改进机器学习方法并更现实地模拟实际的生物利用度。这也是一种很好的方法，可以将选定的新兴POCs的风险评估优先考虑到没有相关知识或标准参考资料的生物群。

Passive sampler devices (PSDs) are often used to monitor longer term occurrence of micropollutants, and mainly hydrophobic organic chemicals (HOCs), in the environment. PSDs are generally composed of a polymeric sorbent and collect solutes by passive diffusion. Recently, PSDs have been used as potential surrogates for fish/invertebrates in bioconcentration studies for HOCs (logKow 4-6)1. More recently, PSDs have been tailored for polar organic chemicals (POCs). For example, pharmaceuticals and personal care products (PPCPs) have a logKow of -1 to 4 and have multiple ionisation states. This makes modelling for PPCPs significantly more challenging. Laboratory based risk assessment for 7,000 pharmaceuticals (excluding their metabolites and transformation products) is impractical and highly costly. There exists an exciting opportunity in this project to develop and apply predictive approaches to prioritise laboratory testing or remove the need to use model organisms entirely. In 2016, we were the first to successfully model and predict POC uptake rate constants (Rs) onto PSDs1. We have also recently used machine learning by itself to predict PPCP bioconcentration in G. pulex with some limited success in a previously BBSRC funded CASE studentship (BB/K501177/1). Therefore, it is now timely for us to extend this knowledge to 3R-type predictive approaches for additional POC bioconcentration studies now incorporating PSDs to improve the machine learning approach and to mimic the actual bioavailability more realistically. This also represents an excellent way to prioritise risk assessment for selected emerging POCs to biota for which no knowledge or standard reference materials exist.