The role of automation in enhancing reproducibility and interoperability of PBPK models Authors: Abdallah Derbalah, Masoud Jamei, Iain Gardner, Armin Sepp Published: 2025 Dear Editor, We would like to commend Domínguez-Romero et al. [1] for their insightful paper, ‘Making PBPK models More Reproducible in Practice’. The authors provide valuable recommendations to enhance the reproducibility of PBPK models, including standardization, transparent sharing, and thorough documentation. The MOCCASIN tool for converting MATLAB-based PBPK models into SBML format will undoubtedly enhance interoperability and wider accessibility of the modelling resources across the scientific community. While these measures are highly recommended, they only fill their purpose if the primary PBPK model is faultless. With models getting larger and more complex all the time, this is far from guaranteed. That is why we believe that an equally important aspect—automation—deserves more emphasis and is long overdue. We argue that, from the first lines of the code generated, the weakest link in the reproducibility and interoperability of PBPK and other complex biochemical models is the human factor. Advanced models can contain hundreds of ordinary differential equations (ODEs) and thousands of algebraic expressions to describe complex kinetics and reactant transport in physiological systems. It is unreasonable to expect these to be coded manually, within acceptable timeframe, without making typographical errors. While syntax errors are easily identified, and unit conversion in dedicated modelling platforms like QSP Designer [2] or SimBiology [3] helps to rectify many more, some will slip through, and are very difficult to spot. Different individuals attempting to build or reproduce complex models can therefore arrive at different (even if repeatable!) models in their hands. PBPK models are particularly prone to such human error due to structural redundancy, where similar compartmental structures and processes are repeated many times across the model. For example, PBPK models may involve multiple organ compartments, each following nearly identical structural framework with only tissue-specific parameter changes. At least two software solutions have been developed to address this shortcoming. The first is PBPKassembler developed by Sepp et al. [4], which is a MATLAB script to automate the assembly of Simbiology-based two-pore PBPK models, based on a standardized Simbiology ‘generic organ’ structure for outlining the generic reaction scheme and worksheet-based tissue- and species-specific parameter values. Both are compiled into fully functional SimBiology model that is available in four different species: mouse, rat, monkey, and human. Another solution has been applied by Liu et al. [5], who successfully used Ubiquity for the same purpose in R. The sheer size and complexity of the resulting models highlight the challenges of manual coding. Such models would be nearly impossible to assemble manually without introducing errors. Automation was critical in this case, minimizing the chance of errors in the construction of complex biochemical reaction schemes reproducibly while minimizing the chances of coding errors, saving weeks or months of time that otherwise would have been spent on model debugging. Given the modular and systematic nature of PBPK models, the field is well positioned to benefit from automation tools that streamline model construction and ensure compatibility with standardized formats. The less need there is for human intervention in building, adapting, converting, or operating complex models, the more reproducible the results are, as has been demonstrated by Galaxy in advanced bioinformatics workflows [6]. We hope this perspective complements the authors’ valuable contributions and stimulates further discussions on improving reproducibility and interoperability in PBPK modelling. Contributor Information Abdallah Derbalah, Certara Predictive Technologies Division, Certara UK Ltd., Level 2-Acero, 1 Concourse Way, Sheffield S1 2BJ, United Kingdom. Masoud Jamei, Certara Predictive Technologies Division, Certara UK Ltd., Level 2-Acero, 1 Concourse Way, Sheffield S1 2BJ, United Kingdom. Iain Gardner, Certara Predictive Technologies Division, Certara UK Ltd., Level 2-Acero, 1 Concourse Way, Sheffield S1 2BJ, United Kingdom. Armin Sepp, Certara Predictive Technologies Division, Certara UK Ltd., Level 2-Acero, 1 Concourse Way, Sheffield S1 2BJ, United Kingdom. Conflict of interest None declared. Funding None declared. References