PKGPT: Expert-Orchestrated Recursive LLM Agent for Automated NONMEM PopPK Modeling with Human Benchmarking
Pharmaceutics · 18.04.2026 · 18 (4) · 501
DOI: 10.3390/pharmaceutics18040501 ↗Kurz gefasst
Fachartikel in der Zeitschrift Pharmaceutics, online erschienen am 18.04.2026. Verfasst von Hoyoung Kwack und 5 weiteren Autor:innen. Beteiligte Einrichtung: Department of Pharmacy and Yonsei Institute of Pharmaceutical Sciences, Yonsei University, Incheon 21983, Republic of Korea. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: the National Research Foundation of Korea.
Abstract (Originalsprache)
Background/Objectives: Population pharmacokinetic (PopPK) modeling in NONMEM requires iterative, expertise-dependent workflows. Naïve zero-shot prompting of general-purpose large language models (LLMs) typically produces NONMEM code that fails to execute. This study introduces PKGPT, a recursive agentic LLM system designed to automate NONMEM-based PopPK model development and benchmarks its performance against human expert models.
Methods
PKGPT, powered by Google’s Gemini 3.0 Flash, embeds pharmacometrics expertise into phase-specific expert-agent prompts orchestrated across five sequential phases: base model establishment, structural diagnostics, overfitting reduction, random-effects optimization, and covariate analysis. The system recursively executes NONMEM, parses outputs, and iteratively refines control streams. PKGPT was evaluated on three public datasets (warfarin, theophylline, and tobramycin) and benchmarked against independently developed human expert models.
Results
PKGPT consistently produced executable, converging NONMEM models across all three datasets. In warfarin, both PKGPT and the human expert selected a one-compartment oral structure (ADVAN2), but the expert achieved a lower OFV (294.41 vs. 484.43) via covariate scaling. In theophylline, PKGPT produced parameter estimates close to the expert solution (Ka = 1.59 vs. 1.46 h−1; CL = 0.0399 vs. 0.0404 L/h/kg). In tobramycin, PKGPT correctly identified a two-compartment structure but produced physiologically implausible peripheral volume estimates (V2 = 149 L vs. expert’s 13.2 L). Across datasets, PKGPT did not identify clinically established covariates, and run-to-run reproducibility was variable.
Conclusions
PKGPT substantially improves the robustness and usability of LLM-generated NONMEM code compared with naïve zero-shot prompting, accelerating model drafting and iterative refinement, but physiological plausibility and clinical interpretability still require a human-in-the-loop oversight.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Pharmacy and Yonsei Institute of Pharmaceutical Sciences, Yonsei University, Incheon 21983, Republic of Korea.
- Interdisciplinary Program in Bioinformatics, Seoul National University, Seoul 08826, Republic of Korea.
- Department of Integrative Biotechnology, College of Pharmacy, Yonsei University, Incheon 21983, Republic of Korea.
- Department of Computer Science, Seoul National University, Seoul 08826, Republic of Korea.
- College of Pharmacy, Jeonbuk National University, Jeonju 54896, Republic of Korea.
- Graduate Program of Industrial Pharmaceutical Science, Yonsei University, Incheon 21983, Republic of Korea.
ORCID-Profile: Hyunseung Kong, Jiwoo Lim, Byoung-Tak Zhang, Jongsung Hahn, Min Jung Chang
Förderung
- the National Research Foundation of Korea
Zitieren
APA
Kwack, H., Kong, H., Lim, J., Zhang, B. T., Hahn, J., & Chang, M. J. (2026). PKGPT: Expert-Orchestrated Recursive LLM Agent for Automated NONMEM PopPK Modeling with Human Benchmarking. Pharmaceutics, 18(4), 501. https://doi.org/10.3390/pharmaceutics18040501
Vancouver
Kwack H, Kong H, Lim J, Zhang BT, Hahn J, Chang MJ. PKGPT: Expert-Orchestrated Recursive LLM Agent for Automated NONMEM PopPK Modeling with Human Benchmarking. Pharmaceutics. 2026;18(4):501. doi:10.3390/pharmaceutics18040501
BibTeX
@article{Kwack2026,
author = {Kwack, Hoyoung and Kong, Hyunseung and Lim, Jiwoo and Zhang, Byoung-Tak and Hahn, Jongsung and Chang, Min Jung},
title = {PKGPT: Expert-Orchestrated Recursive LLM Agent for Automated NONMEM PopPK Modeling with Human Benchmarking},
journal = {Pharmaceutics},
year = {2026},
volume = {18},
number = {4},
pages = {501},
doi = {10.3390/pharmaceutics18040501}
}
EUVIMED