Nicolas Ratto1, Jean-Baptiste Gourlet1, Matthieu Coudron1, Jean Ponchon1, Nicolas Duteil1, Jérémy Villard1, Loïc Etheve1, Claudio Monteiro1
1Nova In Silico
Introduction Many valuable pharmacometric (PMx) models exist only in unstructured formats—such as PDFs or proprietary source files—making them difficult to maintain, share, or integrate with emerging quantitative systems pharmacology (QSP) frameworks. This fragmentation limits reproducibility and the ability to leverage existing PMx models for broader in silico analyses [1]. Our work addresses this challenge by integrating AI-driven tools into the Jinko platform, thereby streamlining the extraction, conversion, and storage of PMx models in a standardized, searchable digital library. Objectives The primary objective is to automate the import of PMx models using AI, enabling their seamless conversion from PDF documents or source code into a standardized format. This process facilitates maintainability and accessibility while ensuring that these models are immediately interoperable with QSP modules. Ultimately, this approach empowers PMx model owners to extract deeper insights from their existing models and paves the way for accelerated drug repositioning and clinical trial simulation before patient recruitment. Methods An automated pipeline was developed that combines optical character recognition and domain-trained natural language processing to extract key model components—such as structural equations, parameter values, and covariate relationships—from diverse sources. This pipeline operates within a declarative and reproducible environment and employs a structured, multi-step approach that fosters reproducibility and auditability while mitigating the opacity of hidden-state models. The extracted information is first converted into a human-readable intermediate format. A multi-step workflow, driven by a Large Language Model (LLM), then transforms this intermediate format into a controlled and structured representation that can be exported into formats such as Systems Biology Markup Language (SBML). The converted models are then made available in the Jinko platform, where they are stored in a centralized, version-controlled library with rich metadata, ensuring ease of retrieval and modification. Several case studies were used to evaluate the pipeline, including the reimplementation of published drug models, and their subsequent integration with an existing QSP model to simulate combined pharmacometric and disease progression outcomes. Results The AI-driven import pipeline successfully reconstructed PMx models with high fidelity, preserving critical components such as compartmental structure, parameter estimates, and interindividual variability terms. Simulated outputs from the imported models closely matched the original published results, confirming the accuracy of the conversion process. Furthermore, integration within the Jinko platform allowed these models to be combined with QSP modules, enabling comprehensive in silico simulations. For instance, linking an imported population PK model with a QSP disease progression framework provided novel insights into dosing strategies and patient subgroup responses. This unified digital library now facilitates model maintainability and promotes collaborative reuse, significantly reducing the time and effort required to reimplement legacy models [2]. Conclusion By automating the extraction and conversion of PMx models into a standardized format within the Jinko platform, our approach significantly enhances model reproducibility, maintainability, and interoperability with QSP frameworks. This innovation not only streamlines the reuse of existing pharmacometric models but also opens the door to advanced in silico analyses—enabling researchers to explore drug repositioning opportunities and refine clinical trial designs before actual patient enrollment. In doing so, the integration of AI with model management and simulation represents a transformative step toward more efficient, data-driven decision-making in drug development.
[1] Cucurull-Sanchez, Lourdes, et al. “Best Practices to Maximize the Use and Reuse of Quantitative and Systems Pharmacology Models: Recommendations From the United Kingdom Quantitative and Systems Pharmacology Network.” CPT: Pharmacometrics & Systems Pharmacology, vol. 8, no. 5, Mar. 2019, pp. 259–72. Portico, Crossref, https://doi.org/10.1002/psp4.12381. [2] Geerts, Hugo, et al. “A Combined Physiologically-based Pharmacokinetic and Quantitative Systems Pharmacology Model for Modeling Amyloid Aggregation in Alzheimer’s Disease.” CPT: Pharmacometrics & Systems Pharmacology, vol. 12, no. 4, Jan. 2023, pp. 444–61. Portico, Crossref, https://doi.org/10.1002/psp4.12912.
Reference: PAGE 33 (2025) Abstr 11583 [www.page-meeting.org/?abstract=11583]
Poster: Methodology – AI/Machine Learning