Ongoing Research · Systems Biology
Data-driven Dynamic Modeling of Signaling Networks in Triple-Negative Breast Cancer
A data-driven modeling framework for inferring kinase signaling dependencies from time-resolved phosphoproteomic data and integrating them into a mechanistic ODE model for dynamic simulation and perturbation analysis.
Theoretical Biophysics · Humboldt-Universität zu Berlin
Ongoing research
Research context
The project originated from my Master's thesis in the Theoretical Biophysics group of Prof. Edda Klipp at Humboldt-Universität zu Berlin and is currently being further developed as part of an ongoing research collaboration.
The broader goal is to develop quantitative models that connect experimentally observed phosphorylation dynamics with the underlying regulatory structure of cellular signaling networks.
From data to dynamics
Reconstructing dynamic signaling behavior from experimental measurements.
Why dynamic modeling?
Cellular signaling pathways are not isolated linear cascades. Their behavior emerges from interacting kinases, feedback, crosstalk and time-dependent changes in phosphorylation.
The modeling framework therefore aims to move beyond a static representation of signaling interactions and reconstruct how network activity evolves over time.
Dependency inference
Estimating how signaling components influence one another.
A known network topology describes which interactions may occur, but not their relative strength. The model therefore introduces dependency parameters that are inferred directly from experimental phosphorylation trajectories.
Kinase → phosphosite
Relative influence of an upstream kinase on the phosphorylation of a target site.
Phosphosite → kinase activity
Contribution of a kinase's phosphorylation state to its effective activity within the network.
Pᵢ(t) = αᵢ⁰ + Σⱼ αⱼⁱ ( βⱼ⁰ + Σₖ βⱼᵏ Pⱼᵏ(t) )
The nested formulation links the phosphorylation state of upstream kinases to their influence on downstream target sites. Parameter estimation is performed under constraints that preserve the intended relative interpretation of these dependencies.
Mechanistic dynamics
From inferred dependencies to a coupled ODE network.
Embedding network activity into kinetics
The dynamic model describes transcription, translation, degradation, phosphorylation and dephosphorylation through a system of ordinary differential equations based on mass-action kinetics.
The inferred signaling dependencies are embedded directly into the phosphorylation terms of the ODE system. In this way, the statistical information extracted from the experimental data becomes part of a mechanistic description of the network dynamics.
Parameter inference
Fitting the data is only part of the problem.
Complementary numerical optimization strategies are used to estimate the constrained dependency parameters from experimental trajectories. Although different optimization formulations can reproduce the data similarly well, they do not necessarily identify the same underlying parameter configuration.
Good agreement with experimental observations does not necessarily imply unique identification of the underlying biological parameters.
This identifiability problem motivates the integration of additional biophysical information and constraints into future versions of the model.
Current work
The model is still evolving.
My current work focuses on improving the kinetic parameter estimation of the dynamic ODE model, with particular attention to numerical stability, computational efficiency and the quality of the resulting simulations.
The modeling framework is intended to remain scalable so that the same underlying approach can be applied to increasingly large and interconnected signaling networks.
In parallel, I am developing the scientific framework for a future publication together with collaborators involved in the experimental research underlying the project.
Toward predictive signaling models
From reconstruction toward prediction and perturbation.
Biophysical constraints
Integrating additional biological knowledge to improve parameter identifiability and mechanistic plausibility.
Improved inference
Developing more robust strategies for estimating kinetic and network parameters from experimental time-series data.
Network perturbations
Using the dynamic model to investigate regulatory control, robustness and responses to targeted in-silico interventions.
Scientific machine learning
Exploring how mechanistic models and modern learning approaches can be combined as larger experimental datasets become available.