A systematic model identification method for chemical transformation pathways \u2013 the case of heroin biomarkers in wastewater

Pedram Ramin,Fabio Polesel,Borja Valverde-Pérez,Luca Locatelli,Benedek Gy Plósz

doi:10.1038/s41598-017-09313-y

Pedram Ramin, Fabio Polesel + Show 3 more

Open Access

https://doi.org/10.1038/s41598-017-09313-y

Copy DOI

Abstract

This study presents a novel statistical approach for identifying sequenced chemical transformation pathways in combination with reaction kinetics models. The proposed method relies on sound uncertainty propagation by considering parameter ranges and associated probability distribution obtained at any given transformation pathway levels as priors for parameter estimation at any subsequent transformation levels. The method was applied to calibrate a model predicting the transformation in untreated wastewater of six biomarkers, excreted following human metabolism of heroin and codeine. The method developed was compared to parameter estimation methods commonly encountered in literature (i.e., estimation of all parameters at the same time and parameter estimation with fix values for upstream parameters) by assessing the model prediction accuracy, parameter identifiability and uncertainty analysis. Results obtained suggest that the method developed has the potential to outperform conventional approaches in terms of prediction accuracy, transformation pathway identification and parameter identifiability. This method can be used in conjunction with optimal experimental designs to effectively identify model structures and parameters. This method can also offer a platform to promote a closer interaction between analytical chemists and modellers to identify models for biochemical transformation pathways, being a prominent example for the emerging field of wastewater-based epidemiology.

Highlights

Models are mathematical representations of real systems that are able to predict their performance under defined conditions
Techniques proposed for uncertainty and identifiability assessment of model parameters include (i) methods based on local analysis (e.g., Brun et al.7), which have been demonstrated to be sensitive to the initial parameter value choice[17, 18]; and (ii) global methods, including Monte Carlo-based (MC) methods such as Generalized Likelihood Uncertainty Estimation (GLUE)[19], or Markov chain- Monte Carlo (MCMC)[20, 21] based methods
Results suggest that Method 1 typically resulted in increased values of linear correlation coefficients (LCC) compared to Method 2, which results from incorporating uncertainties propagated from other upstream model parameters

Summary

Introduction

Models are mathematical representations of real systems that are able to predict their performance under defined conditions. Techniques proposed for uncertainty and identifiability assessment of model parameters include (i) methods based on local analysis (e.g., Brun et al.7), which have been demonstrated to be sensitive to the initial parameter value choice[17, 18]; and (ii) global methods, including Monte Carlo-based (MC) methods such as Generalized Likelihood Uncertainty Estimation (GLUE)[19], or Markov chain- Monte Carlo (MCMC)[20, 21] based methods. It is suggested to assess the impact of parameter uncertainty on model outputs. This task can be carried out via MC simulations[22, 23]. The propagation of uncertainty between model parameters is not considered in any calibration protocols[12, 24, 25]

Objectives

Methods

Results