Biochemical interactions play a crucial role in human exposure to various substances, through food, water, cosmetics, agrochemicals and other consumer products, and provide key information for performing complete toxicological risk-assessments. Unfortunately, many of these substances lack comprehensive toxicological data. To address some of these data gaps, the biochemical characterization of these substances can be achieved by integrating in vitro bioactivity data from high-throughput screening efforts with physiologically based toxicokinetic (PBTK) modelling, facilitating in vitro to in vivo extrapolation (IVIVE).

Open-source computational workflows for PBTK models, such as TKPlate, can be utilized by scientists and regulators without coding expertise, but these are often limited to single compounds. To facilitate high-throughput screening of substances, we developed a KNIME workflow that compiles the necessary compound-specific physicochemical and ADME parameters for a selected list of substances. These can then be processed through PBTK models to obtain in vivo dose equivalents for in vitro assays.

As a proof of concept for this workflow, we selected a list of publicly available substances, which then was used to batch search the US EPA CompTox Chemicals Dashboard to obtain QSAR-ready SMILES and OPERA Model predictions. This output was used as the input to run the oral, inhalation, and buccal PBTK models in both rat and human using the httk R package. Running the KNIME workflow on the approximately 2,500 substances resulted in the generation of predicted Cmax for each compartment specified in parameters, the plasma AUC, and the half-life of the substances for each model.

Comparisons of predicted plasma Cmax for the three different models for both species provide evidence for animal to human extrapolation which could be used in toxicological risk assessments, such as estimating derived human NOAEL from rat POD data. These results also provide insight into which compounds fall outside of the applicability domain for the different models.

The use of this PBTK workflow allows for the standardization of a method to predict in vivo dose information to assess the potential toxicological risk of large sets of substances with limited data in a reproducible and user-friendly way for scientist without coding expertise.

Alexandre Borrel

Appearances