Mahmoud Ali afifi 1, Felix Boakye-Agyeman 1
1 Certara (, United States)
Objectives
Pharmacometric analyses increasingly rely on complex nonlinear mixed-effects (NLME) modeling, large-scale simulations, and simulation-based inference to support study design, regulatory submissions, and critical decision-making. However, practical implementation of such workflows is frequently constrained by computational burden, limited scalability, and challenges in ensuring full reproducibility.
The NLME engine NLME and its R interface RsNLME provide an integrated modeling and simulation environment designed to address these limitations (1).
The primary objective of this work was to demonstrate the capability of the RsNLME framework to execute complex, multi-dimensional pharmacometric (PMX) analyses efficiently, reproducibly, and at industrial scale. Specifically, we implemented population pharmacokinetic (PK) modeling, extensive virtual clinical trial simulations, repeated high-throughput model fitting, and comprehensive power evaluations across diverse and realistic study scenarios.
Methodology
An automated and computationally intensive workflow was developed using the RsNLME framework to evaluate the robustness of Model-Based Bioequivalence (MBBE) compared with Non-Compartmental Analysis (NCA).
Step 1: Master Dataset Simulation
An extensive database was simulated for trastuzumab and clenoliximab using published population PK models we call these models as “True Model”.
· N = 1000 subjects
· Very rich sampling schedule
· Designed to establish the absolute pharmacokinetic truth
· Covered all bioequivalence scenarios that we want to test.
Step 2: Scenario Design and Filtering
A total of 297 unique scenarios per drug were generated by filtering the master dataset according to:
· Sample size (N-subjects): Targeted to achieve 90% power
· Relative bioavailability (Frel): 1.05, 1.1, and 1.25
· Sampling schedules: Optimized using PopED and categorized as rich, moderate, or sparse
Step 3: Automated Workflow and Analyst Management
Execution was managed through a fully standardized automated R script:
· Automatic detection of the analyst running each batch
· Automatic generation of well-structured, scenario-specific folder hierarchies
· Seamless upload of results to Microsoft SharePoint
· Real-time progress monitoring across analysts and scenarios
This ensured data integrity, reproducibility, and transparency in a multi-user environment.
Step 4: Bootstrap Replicate Generation
For each of the 297 scenarios, 500 bootstrap replicates were generated to ensure robust statistical power for bioequivalence assessment.
Step 5: Three-Branch Estimation Scenarios
For every replicate, the automated script executed three-branch estimation of PK metrics:
1. Classical NCA – Traditional non-compartmental analysis
2. True Model – Fitting the original data-generating model
3. Intentionally Misspecified Models – Stress-testing MBBE by:
o Reducing structural complexity (e.g., 3-compartment → 2-compartment)
o Simplifying elimination pathways (e.g., mixed linear/nonlinear → nonlinear only)
Step 6: Bioequivalence Evaluation
Bioequivalence (BE) was assessed for all replicates and estimation branches using the Two One-Sided Tests (TOST) procedure.
Step 7: Automated Data Consolidation
The workflow automatically processed a massive volume of simulation output:
· 148,500 model fits per drug
· Automated generation of standardized summary tables
· Automated construction of final power curves
· High-performance computing using 18 virtual machines (736 cores)
Results
Massive Computational Throughput
The RsNLME framework demonstrated its ability to manage extremely high computational demand:
· Scale: >148,500 model fits per drug
· Scenario complexity: 297 scenarios × 500 replicates
· Infrastructure: 18 virtual machines, 736 cores
· Parallel processing: Thousands of simultaneous simulations and estimations
Automated Operational Intelligence
The standardized automation framework provided:
· Analyst tracking: Automatic user identification
· Real-time monitoring: SharePoint-integrated progress updates
· Systematic data architecture: Scenario-based structured folders
· End-to-end execution: From population simulation to final power curves without manual intervention
This demonstrates that RsNLME functions not only as a modeling tool, but as an industrial-scale computational engine for high-density pharmacometric analysis.
Conclusion
The RsNLME framework enables the execution of high-throughput pharmacometric workflows with reproducible results. It integrates the established NLME modeling engine with automated workflow management, allowing:
· Execution of over 148,500 model fits per drug across 297 scenarios
· Distributed computation using 18 virtual machines with 736 cores
· Structured, scenario-specific folder organization
· Analyst-specific tracking and centralized progress monitoring via SharePoint
The workflow supports population PK modeling, large-scale simulations, and power evaluations with minimal manual intervention. These results demonstrate that RsNLME can perform large-scale, multi-scenario pharmacometric analyses efficiently and reproducibly.
References:
Chen R, Sale M, Craig J, Tomashevskiy M, Mazur A, Hu S, Nieforth K. Pirana and Integrated PMX Tools, a Workbench for NONMEM, NLME, pyDarwin, and RsNLME. CPT Pharmacometrics Syst Pharmacol. 2025 Aug;14(8):1298-1309. doi: 10.1002/psp4.70067. Epub 2025 Jul 4. PMID: 40613737; PMCID: PMC12358299.
Reference: PAGE 34 (2026) Abstr 12120 [www.page-meeting.org/?abstract=12120]
Poster: Software Demonstration