Venetia Karamitsou1, Raphaëlle Lesage1, Rudolf Engelke1, Marco Siccardi1, Stephan Schaller1, Alexander Kulesza1
1ESQlabs GmbH
Introduction: Physiologically Based Pharmacokinetic (PBPK) and Quantitative Systems Pharmacology (QSP) models have become increasingly valuable in drug development ([1], [2]). One critical application is in the modeling of Drug-Induced Liver Injury (DILI), a major cause of drug attrition and regulatory concern. Traditional methods for assessing DILI risk include animal studies and clinical trials and face ethical, logistical, and financial limitations, particularly in identifying rare adverse events in small preclinical/clinical trial populations. Existing computational approaches ([3],[4],[5]) are empirical and mechanistic but are not open-source, lack explainability or cannot cover the often not clearly dose-dependent idiosyncratic nature of DILI [6]. Large virtual populations (10,000+ patients) enable the systematic evaluation of variability in drug response across diverse patient profiles, however, generating them efficiently is a computational challenge. We present an endeavor to build an open-source DILI model and a scalable workflow using R and the Open Systems Pharmacology (OSP) Suite [7] to generate a large virtual dog population that reproduces three major DILI phenotypes, including realistic incidence rates. Our work is conducted as part of the CRACK IT Virtual Second Species Challenge [8], an initiative by the UK NC3Rs (National Centre for the Replacement, Refinement, and Reduction of Animals in Research) aimed at reducing the reliance on dog studies in drug development. Our approach integrates PBPK modeling with a QSP effect model, providing a mechanistic representation of drug distribution, metabolism, and toxicity. Furthermore, we utilize machine learning and cloud-based simulations to significantly enhance computational efficiency and enable the rapid generation of biologically diverse canine virtual patients (VPs). Methods: Sobol Global Sensitivity Analysis is used to identify the most influential QSP model parameters affecting key liver biomarkers (ALT, AST, ALP, necrosis, and liver fat percentage). Latin Hypercube Sampling is then used to generate a diverse set of candidate canine VPs by ensuring uniform coverage of biologically plausible parameter values. Each candidate VP is tested for biological realism by simulating relevant baseline biomarker levels at steady state and rejecting any VP outside predefined reference ranges . Plausible VPs are then simulated under documented drug regimens known to induce specific DILI phenotypes: Acetaminophen (IV, 5g) for acute liver failure, Bosentan (IV, 400mg) for cholestasis, and a chronic Amiodarone treatment (oral, 100mg q12h for 7 days, then 50mg q24h for 1 year) for steatosis, with the goal of reproducing real-world incidence rates for each phenotype. Given the rarity of steatosis, Bayesian Optimization is employed as a machine learning method to efficiently generate high-risk VPs. The workflow is parallelized and deployed on cloud computing infrastructure to enable fast execution of the PBPK- QSP effect model simulations, rapid virtual patient generation, and scalable analysis. Results: The generated virtual dog population exhibited substantial inter-individual variability in liver biomarker levels both at steady state and in treatment response, reflecting real-world diversity in DILI susceptibility. At baseline (in IU/L): •ALT – median: 27, IQR: 15-52, reference normal range: 12-118 •AST – median: 20, IQR: 17-33, reference normal range: 15-66 •ALP – median: 51, IQR: 18-93, reference normal range: 5-131 When simulating drug exposure, the virtual dog population reproduced observed inter-individual differences in DILI outcomes, with Acetaminophen overdose-induced necrosis severity, Bosentan-induced cholestasis incidence, and Amiodarone-induced steatosis rates aligning with published data. Conclusion: Our DILI virtual population generation pipeline is part of an open-source initiative that will be integrated into the CRACK IT Virtual Dog Suite based on R and the OSP Suite. It integrates PBPK and QSP modeling, machine-learning enriched sampling of parameters and cloud infrastructure resources, enabling efficient and robust DILI risk assessment in dog that accounts for inter-patient variability and idiosyncratic reactions along with dose-effect responses, while maintaining full transparency and reproducibility.
[1] Isoherranen, N. (2024). Physiologically Based Pharmacokinetic (PBPK) Modeling of small molecules: How Much Progress Have We Made? Drug Metabolism and Disposition, 53(1), 100013. https://doi.org/10.1124/dmd.123.000960 [2] Schoeberl, B., Musante, C.J., Ramanujan, S. (2024). Future Directions for Quantitative Systems Pharmacology. In: Handbook of Experimental Pharmacology. Springer, Berlin, Heidelberg. https://doi.org/10.1007/164_2024_737 [3] DILIsym. (2025). DILIsym® Software [Computer software]. Simulations Plus, Inc. Retrieved from https://www.simulations-plus.com/software/dilisym/ [4] Li, T., Tong, W., Roberts, R., Liu, Z., & Thakkar, S. (2020). DEEPDILI: Deep Learning-Powered Drug-Induced Liver Injury Prediction using Model-Level Representation. Chemical Research in Toxicology, 34(2), 550–565. https://doi.org/10.1021/acs.chemrestox.0c00374 [5] Seal, S., Williams, D., Hosseini-Gerami, L., Mahale, M., Carpenter, A. E., Spjuth, O., & Bender, A. (2024). Improved detection of Drug-Induced liver injury by integrating predicted in vivo and in vitro data. Chemical Research in Toxicology, 37(8), 1290–1305. https://doi.org/10.1021/acs.chemrestox.4c00015 [6] Hosack, T., Damry, D., & Biswas, S. (2023). Drug-induced liver injury: a comprehensive review. Therapeutic Advances in Gastroenterology, 16. https://doi.org/10.1177/17562848231163410 [7] Open Systems Pharmacology Suite (OSPS) v12, https://github.com/Open-Systems-Pharmacology/Suite [8] CRACK IT Virtual Second Species Challenge, https://nc3rs.org.uk/crackit/virtual-second-species
Reference: PAGE 33 (2025) Abstr 11491 [www.page-meeting.org/?abstract=11491]
Poster: Methodology - New Modelling Approaches