In human health risk assessment, pediatric physiologically-based pharmacokinetic modeling has been used to predict chemical-specific toxicokinetic adjustment factors. An important input parameter fraction unbound in plasma in child (fupchild) can be scaled from fupadult using ontogeny models. Multiple computational tools can be used to predict fupchild for data-poor compounds but prediction accuracy can be different depending on data-availability. We evaluated prediction of fupchild for data-limited scenario with data-rich compounds. The objectives were (i) to evaluate the uncertainty of the ontogeny models in different data availability scenarios and (ii) to evaluate how these different data availabilities can impact the overall prediction accuracy of fupchild. Six calculation methods that emulate various data-availability scenarios were applied to calculate 139 fupchild values. The ionization state at pH 7.4 and QSPR-predicted fupadult values were calculated from the structure of a molecule using ADMET Predictor. When the albumin ontogeny equation was applied to calculate fupchild for all compounds, fupchild values were under-predicted compared to observed fupchild with average fold error (AFE) values from 0.68 to 0.74. When the AAG model was used, a less degree of bias was observed. When the binding partner was assumed based on acid-base properties, the prediction accuracy was similar to the cases where fupchild values were calculated based on known binding partner information. When QSPR-predicted fupadult values were used as an input, substantial over-predictions were observed for acids and neutrals with an AFE up to 8. The results demonstrated that an experimental determination of fupadult was crucial and prediction of acidic or basic ionization status at pH 7.4 was sufficient to select the most appropriate ontogeny model that could predict fupchild from fupadult.
By Yejin Esther Yun, Andrea N. Edginton