Diane Lefaudeux1, Rudolf Engelke1, Pavel Balazki1, Marco Siccardi1, Stephan Schaller1
1ESQlabs GmbH
Introduction The increasing importance of Physiologically-Based Pharmacokinetic (PBPK) models in drug development and chemical risk assessment highlights the growing need to effectively use PBPK models across a high number of compounds. Our goal is to develop a robust framework in R that facilitates simulating a high number of PBPK models integrating in vitro and in silico compound data in an efficient manner using the open-source Open Systems Pharmacology (OSP) [1] ecosystem. To this end, we developed a framework to run PBPK models in a high-throughput and user-friendly approach, testing it on a sample set of anti-tuberculosis drugs. Materials and methods Using open-source R packages from the OSP ecosystem, we developed a framework that allows efficient configuration and simulations of many compounds studies by leveraging the various packages capabilities. Results Using the capabilities of the open-source R packages from the OSP ecosystem, we developed a framework to predict the in-vivo PK profile of many drugs or chemicals. To this end, we leveraged the possibility of reusing the same model structure to support model reproducibility. The framework allows PK predictions in the various species (human, rat, mouse, monkey, dog, …) using different types of biological processes (linear hepatic clearance, renal clearance, biliary clearance) as available in PK-Sim. To enhance usability, the pipeline can automatically generate a generic model incorporating the required species and processes based on defined compound properties (e.g., physicochemical characteristics, clearance mechanisms) and study design (e.g., species, administration protocol). This is achieved using any PK-Sim default individuals and processes. The approach also supports the flexible integration of pharmacodynamic effects from a MoBi model. Once the generic models are established, simulations are run in parallel by systematically varying key parameters—including physicochemical properties, biological processes (e.g., clearance values), and administration protocol details (e.g., dose, route, duration, frequency). The pipeline then processes the defined study conditions and returns the resulting PK profiles, enabling efficient and scalable simulations across a range of compounds and datasets. In general, PBPK models parametrized with physico-chemical properties and in-silico / in vitro data only allowed predictions of in vivo PK profiles with AUC within 2-fold 40% of the time [3]. We tested the framework on a sample set of 12 anti-tuberculosis drugs that included a vast variety of administration protocols (single dose intravenous to multiple oral doses with loading doses), representing a total of over 100 different studies (different dose regimens and drugs) extracted from literature to ensure its usability and flexibility using QSAR predictions as inputs. On this initial set of studies focused on anti-TB drugs, we observed a simulated AUC and Cmax within 2-fold of the observed values for 38-42% across all studies (AUC within 4-fold ~60% of the studies; Cmax with 4-fold in ~75% of the studies), this seems to be in agreement with previously observed values on different datasets. The exact number varies slightly depending on the partition coefficient method. The median AUC ratio across all studies is 0.6-fold, while the median Cmax ratio is 1.3-fold. Conclusion An efficient HT-PBPK framework has broad applications and can serve as a valuable tool to streamline drug development. The development framework can integrate efficiently QSAR and IVIVE, providing quantitative PK estimates where little to no in vivo data is available. Additionally, it can be integrated with machine learning approaches to refine input parameter predictions for PBPK models [2]. The framework can facilitate retrospective analysis on large databases of compounds with available in vivo PK data, allowing systematic evaluation of different modeling methodologies and structural model hypotheses [3]. Moreover, it can help optimise and calibrate IVIVE workflows, such as refining intestinal permeability estimates from in vitro assays, and efficiently parametrizing PBPK models for multiple compounds when in vivo PK data is available. This approach has demonstrated innovative impact, with new opportunities to streamline anti-TB drug development by accelerating data-driven decision-making and improving predictive performance in the early-stage development of new compounds
[1] Open Systems Pharmacology Community. Open Systems Pharmacology [Internet]. [cited 2024 Aug 28]. Available from: https://www.open-systems-pharmacology.org/ [2] Walter M, Aljayyoussi G, Gerner B, Rapp H, Tautermann CS, Balazki P, et al. In silico PK predictions in Drug Discovery: Benchmarking of Strategies to Integrate Machine Learning with Empiric and Mechanistic PK modelling [Internet]. bioRxiv; 2024 [cited 2024 Aug 29]. p. 2024.07.30.605777. Available from: https://www.biorxiv.org/content/10.1101/2024.07.30.605777v1 [3] Geci R, Gadaleta D, de Lomana MG, Ortega-Vallbona R, Colombo E, Serrano-Candelas E, et al. Systematic evaluation of high-throughput PBK modelling strategies for the prediction of intravenous and oral pharmacokinetics in humans. Arch Toxicol. 2024;98(8):2659–76.
Reference: PAGE 33 (2025) Abstr 11529 [www.page-meeting.org/?abstract=11529]
Poster: Methodology - Other topics