Pumas: A Unified Platform for Pharmaceutical Modeling and Simulation

ABSTRACT Pumas (Pharmaceutical Modeling and Simulation) is a software ecosystem for quantitative pharmaceutical analytics built on the Julia programming language. The use of “pharmaceutical modeling” is deliberate. Pumas is designed as a unified platform for quantitative analytics across the full horizontal of drug development. The classical pharmacometrics segment; non-compartmental analysis, population PK/PD, quantitative systems pharmacology, physiologically based pharmacokinetic modeling, bioequivalence, and in-silico clinical trials represents one part of this continuum. Pumas extends this to machine learning–augmented pharmacology (DeepPumas), epidemiological modeling, pharmacoeconomics, quality-by-design, clinical trial design and optimization, precision dosing, real-world evidence analytics, and many more. While only a subset of these capabilities has been released to date, the architectural vision is to serve the analytical needs of pharmaceutical R&D end-to-end, from target identification through regulatory submission and post-market surveillance. The term “pharmaceutical modeling” reflects this scope: not pharmacometrics alone, but the entirety of quantitative science that underpins modern drug development within a single scientific computing ecosystem. This article introduces the Pumas ecosystem and demonstrates its integrated workflow through a complete case study of a multiple ascending dose trial with pharmacokinetic (PK) and pharmacodynamic (PD) endpoints. At this moment, the platform integrates many related capability domains: Clinical Data Interchange Standards Consortium (CDISC)-compliant data preparation, exploratory data analysis with publication-quality visualization, non-compartmental analysis (NCA) with bioequivalence testing, nonlinear mixed effects (NLME) modeling with multiple estimation algorithms (e.g., First Order Conditional Estimation (FOCE), Laplacian, Stochastic Approximation Expectation Maximization (SAEM), Monte Carlo Expectation Maximization (MCEM), Bayesian), simulation with differential equations using a variety of solvers, machine learning integration via DeepPumas, optimal experimental design, parallel computing infrastructure, and automated reporting through Quarto integration. We demonstrate the unified workflow from data preparation through model-based dose selection. The example use case is chosen to be reasonably simple to focus on the workflow itself yet realistic enough to illustrate key concepts. Using a case study with 60 subjects across 6 dose levels, we develop a two-compartment pharmacokinetic model with covariate effects (sex on clearance, weight on volume), extend to pharmacodynamics with an indirect response model, and simulate alternative dosing regimens to evaluate target attainment metrics. The complete analysis executes within a single computational framework, eliminating tool-switching overhead and data conversion steps. Pumas’s design principles include multiple dispatch, composability, and performance that enable both rigorous statistical modeling and rapid iterative exploration. Integration across workflow stages reduces friction in pharmacometric analyses, supporting quantitative decision-making from first-in-human studies through regulatory submission. The platform supports Phase 1–3 clinical trial analysis and model-informed drug development applications.