PhysiBoSS 2.0: a sustainable integration of stochastic Boolean and agent-based modelling frameworks

Systems biology heavily relies on mathematical models and simulations to understand complex biological systems. Different frameworks exist, such as Boolean modeling for signaling networks and agent-based modeling (ABM) for multicellular systems. Boolean models use logical rules and discrete time to simulate signal propagation and phenotypic changes, frequently used in cancer research for understanding signaling networks and personalizing treatment. ABM simulates cell populations and their interactions with the environment, finding use in fields like microbial ecology and cancer research. Multi-scale models, like PhysiCell, integrate solvers for various processes (diffusion, mechanics, cell growth) at different scales. PhysiBoSS 1.0 was an early attempt to combine PhysiCell and MaBOSS (a Boolean model simulator), but it suffered from maintainability issues due to its design. PhysiBoSS 2.0 addresses these issues.

The introduction reviews existing literature on Boolean modeling in signaling networks and agent-based modeling in multicellular systems, highlighting their applications in cancer research. It mentions PhysiCell, a multi-scale ABM framework, and its previous extension, PhysiBoSS 1.0. The limitations of PhysiBoSS 1.0 in terms of maintainability and flexibility are discussed, setting the stage for the introduction of PhysiBoSS 2.0.

PhysiBoSS 2.0 is implemented as a decoupled add-on to PhysiCell, allowing independent upgrades of both components. It incorporates features like customizable Boolean models and cell types defined in XML configuration files. The PhysiCell Tool Kit (PCTK), a Python package, is introduced for handling and visualizing simulation outputs. The methodology includes re-implementing models from PhysiBoSS 1.0 to validate the new version. Two examples are presented: integration of cell receptor models coupling environmental signals to Boolean models, and integration of pharmacodynamics and Boolean models for in silico drug screening. The drug screening involves using dose-response curves from the Genomics of Drug Sensitivity in Cancer (GDSC) database to model drug effects on specific nodes in the Boolean model. The Bliss independence model is used to assess drug synergies.

PhysiBoSS 2.0 successfully reproduces results from PhysiBoSS 1.0, demonstrating its validity. The new design enhances modularity and reusability. The authors showcase drug screening studies on a prostate cancer cell line (LNCaP) using six drugs. Two drugs (Ipatasertib and Pictilisib) showed significant growth inhibition. Drug combinations also yielded significant results, showing synergies between Pictilisib and Ipatasertib, and Pictilisib and Luminespib. The analysis reveals complex synergies depending on drug concentrations, highlighting the importance of considering concentration-dependent effects. The simulations also explored heterogeneity in drug response due to uneven drug penetration in the 3D spheroid, showing that outer layers can shield inner layers, but death of outer-layer cells can increase space for proliferation in inner layers.

PhysiBoSS 2.0's modular design improves maintainability and allows for easier extension with new models. The drug screening results demonstrate the potential of the framework for studying drug synergies and tumor heterogeneity. The observed heterogeneity in drug response, due to both uneven drug penetration and genetic differences, highlights the importance of considering these factors in developing effective cancer therapies. The study provides insights into the complex interplay between drug concentration, spatial location within the tumor, and cell fate.

PhysiBoSS 2.0 offers a significant improvement over its predecessor by providing a more sustainable, flexible, and extensible framework for multi-scale modeling. The successful reproduction of previous results and the demonstration of novel applications in drug screening and heterogeneity studies underscore its value for systems biology research. Future work will focus on expanding the model to include extracellular matrix, blood vessels, and complex 3D architectures derived from spatial omics data.

The study uses simplified models of drug interactions and tumor microenvironment. The in silico drug screening, while informative, may not fully capture the complexity of in vivo drug responses. The model's accuracy depends on the quality and completeness of the input data, including the Boolean models and dose-response curves. While validated against some experimental data, further validation is necessary to fully assess the model's predictive power.