Characterizing Accuracy of Model Predictions for Chemical Concentration in High Throughput Screening Assays

The aim of in vitro - in vivo extrapolation (IVIVE) of dose is to predict potential whole animal effects from perturbations observed in vitro. Reliance on nominal (administered) concentration (assumed dose) does not account for chemical distribution to cellular and noncellular elements of an in vitro assay system. Chemical distribution can reduce the free concentration available to cause effects and can result in an inaccurate estimate of the intracellular concentration causing any observed perturbations. There are mathematical, chemical property-based partitioning models for predicting cellular concentrations when not measured experimentally. This work evaluated two of these in vitro disposition models, Armitage et al., 2021 and Kramer et al., 2010, using a total of 153 experimental intracellular concentration measurements of 43 chemicals, along with parameters describing measurement conditions, from 12 peer reviewed studies in addition to data generated for this study. Intracellular concentrations were more accurately predicted by both the Armitage model (root mean squared log₁₀ error (RMSLE) = 1.12) and the Kramer model (RMSLE = 1.30) than by the nominal concentration (RMSLE = 1.45). Although limited by the amount of available measurement data, these results indicate that mathematical modeling of in vitro distribution can improve the accuracy of IVIVE for toxicology.