Leveraging Open Systems Biology Standards for Improved Reusability and Reproducibility of Pharmacological Models
Paul Lang (1), Sebastian Miclu?a-Câmpeanu (1), Elisabeth Roesch (1), Chris Rackauckas (1,2,3)
JuliaHub (1), PumasAI (2), Computer Science and Artificial Intelligence Laboratory (CSAIL), Massachusetts Institute of Technology (3)
Introduction/Objectives:
In systems biology, open standards for model exchange and storage have been designed to enable model transfer between computing platforms and to ensure that model descriptions can outlast the software used to create them. These standards provide a structured and modular approach to thinking about models, thus facilitating model reusability and reproducibility. This is especially important in long-term multi-team modeling efforts in the pharmacological industry. However, standards are less flexible than barebones computer code. Here, we set out to identify where these standards can help in the development process of pharmacological models and where they fall short compared to dedicated pharmacology standards. In particular, we focus on the ability of open standards to (1) describe biology with a mathematical model, (2) specify experimental readouts, (2) define dosing schemes and trials, (4) specify QSP and (5) NLME optimization problems.
Methods:
We used the quantitative systems pharmacology tool PumasQSP, which has import capability for open standards such as BioNetGen [1, 2], SBML [3], CellML [4], and PEtab [5]. To illustrate the potential and limitations of these standards, we took a model of insulin and IGF1 signaling [6] through the PumasQSP modeling workflow.
Results:
- Defining biological mechanisms. The insulin/IGF1 signaling model was specified in BioNetGen, which provides a particularly structured, abstract, and intuitive way to describe biochemical processes. BioNetGen can be exported to SBML and used for both deterministic and stochastic simulations.
- Defining experimental readouts. The PEtab format allows for the independent specification of experimental readouts in tabular form, including a noise model with known or unknown parameters.
- Defining dosing schemes and trials. Dosing schemes are limited to setting initial conditions or parameters such as influx rates at the beginning of each trial. Changing these mid-trial either requires to split and chain multiple trials together, or to introduce events and corresponding parameters into the biological model, i.e. the BioNetGen, SBML or CellML file. Both violate the principle of modularity between specification of biological mechanisms and experimental/therapeutic interventions.
- Defining QSP optimization problems. PEtab allows to define priors and search windows for the parameters in the biological model, the observation model (including noise) and the trial condition (e.g. to estimate unknown side effects). This, together with the capability to specify time course measurements, compatibility with SBML and BioNetGen, and the definition of a noise model (2), defines statistically grounded likelihood or posterior probabilities as QSP optimisation objectives. Using PumasQSP and the Insulin/IGF1 signaling model, we demonstrate how to generate a virtual population from a PEtab optimization problem.
- Defining NLME optimization problems. At present the investigated open systems biology standards lack the ability to define mixed effects. For NLME models, dedicated pharmacology standards such as PharmML [7] are therefore more suitable.
Conclusion:
Open standards have the potential to facilitate the definition of models and optimization problems in the pharmacological domain. Their structured and modular approach can improve model communication between humans and machines, leading to greater reusability and reproducibility. PEtab’s tabular schema is particularly easy to parse for humans. However, further extensions are needed to enable the definition of dosing regimens and mixed effects, in order to meet the requirements of the industry.
References:
[1] Michael L. Blinov et al. “BioNetGen: software for rule-based modeling of signal transduction based on the interactions of molecular domains”. Bioinformatics (2004).
[2] Leonard A. Harris et al. “BioNetGen 2.2: advances in rule-based modeling”.
Bioinformatics (2016).
[3] M. Hucka et al. “The systems biology markup language (SBML): a medium
for representation and exchange of biochemical network models”. Bioinformatics (2003).
[4] Autumn A. Cuellar et al. “An Overview of CellML 1.1, a Biological Model
Description Language”. SIMULATION (2003).
[5] Leonard Schmiester et al. “PEtab—Interoperable specification of parameter estimation problems in systems biology”. PLOS Computational Biology (2021).
[6] Cemal Erdem et al. “Inhibition of RPS6K reveals context-dependent Akt activity in luminal breast cancer cells”. PLOS Computational Biology (2021).
[7] Roberto Bizzotto et al. “PharmML in Action: an Interoperable Language for Modeling and Simulation”. CPT Pharmacometrics Syst Pharmacol (2017).