The First Attempt at Non-Linear in Silico Prediction of Sampling Rates for Polar Organic Chemical Integrative Samplers (POCIS)

Abstract

Modeling and prediction of polar organic chemical integrative sampler (POCIS) sampling rates (Rs) for 73 compounds using artificial neural networks (ANNs) is presented for the first time. Two models were constructed: the first was developed ab initio using a genetic algorithm (GSD-model) to shortlist 24 descriptors covering constitutional, topological, geometrical and physicochemical properties and the second model was adapted for Rs prediction from a previous chromatographic retention model (RTD-model). Mechanistic evaluation of descriptors showed that models did not require comprehensive a priori information to predict Rs. Average predicted errors for the verification and blind test sets were 0.03 ± 0.02 L d–1 (RTD-model) and 0.03 ± 0.03 L d–1 (GSD-model) relative to experimentally determined Rs. Prediction variability in replicated models was the same or less than for measured Rs. Networks were externally validated using a measured Rs data set of six benzodiazepines. The RTD-model performed best in comparison to the GSD-model for these compounds (average absolute errors of 0.0145 ± 0.008 L d–1 and 0.0437 ± 0.02 L d–1, respectively). Improvements to generalizability of modeling approaches will be reliant on the need for standardized guidelines for Rs measurement. The use of in silico tools for Rs determination represents a more economical approach than laboratory calibrations.

Quantitative structure-property relationships for predicting sorption of pharmaceuticals to sewage sludge during waste water treatment processes

“…This sensitivity is likely due to the diversity of K d measurements available in the literature used to compile the dataset. This effect is often observed when modelling environmental data obtained from multiple sources (Miller et al, 2016). There are few published multivariate models for comparison, apart from the work of Sathyamoorthy and Ramsburg (2013) based on 15 molecular properties (similar to the MOE descriptors) relating to sorption processes.…”

“…Bayesian regularized artificial neural networks (ANNs) were used with these descriptor sets to derive non-linear models. ANNs have been used to obtain predictive models of complex environmental data sets (Barron et al, 2009, Miller et al, 2016). The ANNs are trained by adjusting their parameters to minimize an error, or cost, function on a subset of training data.…”

Experimental Estimation of 44 Pharmaceutical Polar Organic Chemical Integrative Sampler Sampling Rates in an Artificial River under Various Flow Conditions

“…A TWAC can be calculated by using POCIS, although a sampling rate value is still needed. Currently, for POCIS, sampling rate values are determined experimentally, yet a major challenge consists of predicting sampling rate values from the characteristics of compounds to overcome the lack of sampling rate reproducibility (as exhibited in Figure 2), which is mainly due to the wide range of experimental conditions employed (Miller et al 2016). Unfortunately, until now, no studies have given an acceptable predictive model between sampling rates and compound characteristics (Li et al 2010b; Bartelt‐Hunt et al 2011; Ahrens et al 2015).…”

“…Calibration experiments spanned a wide range of conditions, some of which were far from being environmental conditions, for example, compound concentrations from 10 to 10 000 ng/L, various matrices (such as wastewater, tap water, ultrapure water, or sea water), temperature and agitation modes adopted to induce water flow velocity (such as magnetic, flow system, quiescent), and the design and volumes of the laboratory‐scale pilot. For example, calibrations may be conducted in a beaker filled with 1 to 4.5 L of water (Alvarez et al 2004; Jones‐Lepp et al 2004; Matthiessen et al 2006; Arditsoglou and Voutsa 2008; Martínez Bueno et al 2009; Kohoutek et al 2010; Bartelt‐Hunt et al 2011; Rujiralai et al 2011; Thomatou et al 2011; Charlestra et al 2012; Amdany et al 2014; Magi et al 2018), or in a small tank filled with 2 to 10 L (Alvarez et al 2007; MacLeod et al 2007; Togola and Budzinski 2007; Li et al 2010a; Bayen et al 2014; Di Carro et al 2014; Metcalfe et al 2014; Martínez Bueno et al 2016; Miller et al 2016), 20 to 50 L (Hernando et al 2005; Zhang et al 2008; Bailly et al 2013; Morin et al 2013; Vallejo et al 2013; Belles et al 2014a; Djomte et al 2018), 50 to 100 L (Mazzella et al 2007; Lissalde et al 2011; Fauvelle et al 2012; Ahrens et al 2015; Poulier et al 2015), or 250 to 1400 L (Harman et al 2008; Belles et al 2014b; Kaserzon et al 2014) of water. Less frequently, calibrations are conducted in laboratory‐scale pilots with channels containing a large volume of water, on the order of 120 L (Li et al 2010b), 480 L (Vermeirssen et al 2012), or as much as 113 000 L (Lotufo et al 2018).…”

Understanding and predicting the diffusivity of organic compounds in polydimethylsiloxane material for passive sampler applications using a simple quantitative structure–property relationship model

“…The unexpected positive coefficient A i for the descriptor TPSA probably also originates from the limited power of predictability of this descriptor. Sensitivity analysis was undertaken by measuring the ratio between 2 R 2 values, 1 determined using all selected descriptors for the final model and 1 determined using the same set of descriptors minus the descriptor for which the sensitivity analysis was performed (Miller et al 2016). This allows one to examine the model improvement provided by a given descriptor.…”