CHAE, J.-w. | Search bioRxiv

bioRxiv2026

Model-Driven Hybrid AI Framework for End-to-End Autonomous Decision-Making in Drug Development

Decision-making in drug development spans heterogeneous stages from molecular design to clinical optimization, yet computer-aided workflows across stages remain fragmented, limiting traceability from evidence-derived clinical questions to simulation scenarios and decision endpoints. We present a model-driven hybrid AI framework for end-to-end decision support that treats PICO (Participants, Intervention, Comparison, Outcome) as a machine-actionable specification to define scenarios and align endpoints across modules. Given a PICO-defined clinical question and a drug SMILES, the framework infers PBPK-ready parameters from structure, generates exposure via mechanistic PBPK simulation, and links exposure outputs to clinic-facing therapeutic drug monitoring (TDM) decision support using an open-source NLME-based Clinical Pharmacokinetic Consultant Service (CPCS). We additionally implement an LLM-based PICO extraction layer to structure clinical abstracts into P/I/C/O elements for systematic evidence ingestion. Results include backbone verification for DTI-guided intrinsic clearance prediction under fold-error bounds and PBPK trajectory plausibility, and external validation of the hybrid AI-PBPK pipeline showing substantial dispersion for several endpoints. An input-controlled comparison between PhysioSim and PK-Sim on N = 9 drugs yields similarly degraded accuracy under identical AI-predicted inputs, suggesting upstream parameter quality as a dominant constraint. For evidence ingestion, prompting-only PICO extraction achieves F1 of 0.687 on EBM-NLP and 0.552 on TB-PICO. For clinical utility, CPCS improves TDM performance over a PKS baseline on phenobarbital and vancomycin, reducing MAPE by approximately 26-60% depending on configuration. Overall, the framework provides a modular, traceable blueprint linking evidence-defined questions to exposure simulation and TDM decision support while preserving mechanistic interpretability. Future work will integrate outcome-aligned time-to-event modeling and federated learning to close the PICO-to-outcome loop and improve parameter inference under multi-site governance.

pharmacology and toxicology↗

bioRxiv2026

Real world data based evaluation of a novel target-mediated drug disposition approximation model

Target-mediated drug disposition (TMDD) models have been widely used to describe nonlinear pharmacokinetic profiles driven by high-affinity, low-capacity drug-target binding. A pTMDD model, derived by applying the Pade approximation of the quasi-steady-state (QSS) model (qTMDD) was previously proposed. Although pTMDD model showed a comparable estimation accuracy while maintaining computational efficiency, further validation in realistic clinical scenarios and comprehensive performance evaluations have been needed to assess its practical applicability. Here, we evaluated the pTMDD model using five clinical datasets and extended the previous study that focused on simulations. Using the full TMDD as a reference, the approximation models were compared in terms of the parameter estimation results (parameter estimates, relative standard error values and model diagnostics) and computational efficiency (estimation and bootstrap runtimes). The pTMDD model, previously validated in simulation settings, also preserved the estimation accuracy while reducing the computation time of the clinical data. Both pTMDD and qTMDD remained close to the full TMDD model, whereas Michaelis-Menten TMDD (mTMDD) model showed substantial discrepancies especially at low doses, including biased estimates for key TMDD-related parameters (e.g., kdeg, kint, krec, and kup) and higher objective function values. Moreover, pTMDD was faster than qTMDD in four of the five cases compared to the full TMDD. The time savings were particularly pronounced for larger datasets, supporting the computational efficiency of pTMDD. Q2PCONV, an R Shiny application that converts NONMEM code from qTMDD to pTMDD, was also developed, thereby making this new approximation more accessible to researchers. The findings support pTMDD as a practical alternative to existing TMDD approximation models. Author SummaryTarget-mediated drug disposition (TMDD) models describe a high-affinity, low-capacity binding between drug and its target. To avoid overparameterization, approximation models have been used. The two primary models are Michaelis-Menten model (mTMDD), which is accurate only at high doses, and Quasi-steady-state (qTMDD), which is accurate in wider ranges but requires longer runtime. We have proposed a new approximation model named pTMDD. Here, we evaluated pTMDD using five real clinical trial datasets to assess its practical usefulness. pTMDD produced parameter estimates closer to those from the full TMDD model and showed lower uncertainty than both the full TMDD and mTMDD models. In terms of computational efficiency, pTMDD reduced estimation time by an average of 11% and bootstrap time by an average of 6% relative to qTMDD across cases. In addition, we also developed an R shiny application to help researchers apply pTMDD in practice. Our work supports pTMDD as a practical and efficient tool for TMDD modeling in drug development.

pharmacology and toxicology↗