Efficient parameter estimation for ODE models of cellular processes using semi-quantitative data

Domagoj Dorešić, Stephan Grein, Jan Hasenauer

Abstract

Quantitative dynamical models facilitate the understanding of biological processes and the prediction of their dynamics. The parameters of these models are commonly estimated from experimental data. Yet, experimental data generated from different techniques do not provide direct information about the state of the system but a non-linear (monotonic) transformation of it. For such semi-quantitative data, when this transformation is unknown, it is not apparent how the model simulations and the experimental data can be compared. Here, we propose a versatile spline-based approach for the integration of a broad spectrum of semi-quantitative data into parameter estimation. We derive analytical formulas for the gradients of the hierarchical objective function and show that this substantially increases the estimation efficiency. Subsequently, we demonstrate that the method allows for the reliable discovery of unknown measurement transformations. Furthermore, we show that this approach can significantly improve the parameter inference based on semi-quantitative data in comparison to available methods. Modelers can easily apply our method by using our implementation in the open-source Python Parameter EStimation TOolbox (pyPESTO).

Introduction: The Challenge of Semi-Quantitative Data

In systems biology, Ordinary Differential Equations (ODEs) are the standard for modeling cellular dynamics. A critical step in making these models predictive is estimating reaction rates and other mechanistic parameters from experimental data. While quantitative data—where measurements are directly proportional to the substance of interest—is ideal, many modern biological techniques produce semi-quantitative results.

Methods such as FRET imaging, optical density, and various staining techniques often exhibit non-linear mappings. For instance, as protein concentration increases, the measured signal might saturate or follow a sigmoidal curve. When this mapping is unknown, standard calibration methods fail because they cannot accurately compare model simulations to real-world observations.

The non-linear mapping $g$ relates underlying biochemical quantities ($y_i$) to observed measurements ($z_i$).

Methods: Spline-Based Hierarchical Optimization

To bridge the gap between model states and semi-quantitative observations, we developed a framework within the pyPESTO toolbox. The method relies on several key innovations:

  • Monotone Splines: We use piece-wise linear splines to approximate the unknown measurement mapping. These splines are constrained to be monotonic, preserving the logical ordering of the data (e.g., higher concentration always leads to a higher signal).
  • Hierarchical Optimization: The estimation is split into two levels. The "outer" problem optimizes mechanistic parameters, while the "inner" problem determines the optimal spline and noise parameters for a given set of mechanistic guesses.
  • Analytical Gradients: By deriving exact formulas for the gradients of the hierarchical objective function using the Envelope Theorem, we eliminate complex dependency terms at the optimum. This mathematical refinement makes the optimization process significantly faster and more stable.
Experimental data often consists of noise-corrupted measurements ($\tilde{z}_i$) derived from the underlying system state.

Key Results and Insights

The method was validated using a toy Michaelis-Menten model and four biological benchmarks (STAT5 signaling, HIV dynamics, the repressilator circuit, and IL13-induced signaling).

  • Efficiency Gains: Providing analytical gradients for the hierarchical problem makes optimization at least twice as efficient as gradient-free alternatives.
  • Mapping Recovery: The spline-based approach successfully reconstructed synthetic non-linear mappings (e.g., saturation and sigmoidal curves) in the FRET and STAT5 models. MCMC sampling showed that the true mappings fell well within the 95% credibility intervals.
  • Parameter Accuracy: In the IL13-induced signaling model (M4), the spline method achieved parameter estimates nearly as accurate as those obtained with perfectly known measurement mappings, significantly outperforming standard linear estimation which introduced high bias.
  • Diagnostic Utility: The estimated splines serve as a diagnostic tool, allowing researchers to visualize the saturation profiles or detection limits of their experimental setups.

Conclusion and Future Work

Our spline-based approach provides a robust, Likelihood-based framework for integrating semi-quantitative data into biological modeling. By combining hierarchical optimization with analytical gradients, we offer a tool that is both computationally efficient and statistically rigorous.

While powerful, the current use of piece-wise linear splines can introduce small approximation errors due to lack of smoothness. Future work will explore smoother functions, such as Beta distribution CDFs, and automated symbolic identification of measurement mappings to further refine model accuracy and interpretability.

Summary updated

Abstract

Abstract Motivation Quantitative dynamical models facilitate the understanding of biological processes and the prediction of their dynamics. The parameters of these models are commonly estimated from experimental data. Yet, experimental data generated from different techniques do not provide direct information about the state of the system but a nonlinear (monotonic) transformation of it. For such semi-quantitative data, when this transformation is unknown, it is not apparent how the model simulations and the experimental data can be compared. Results We propose a versatile spline-based approach for the integration of a broad spectrum of semi-quantitative data into parameter estimation. We derive analytical formulas for the gradients of the hierarchical objective function and show that this substantially increases the estimation efficiency. Subsequently, we demonstrate that the method allows for the reliable discovery of unknown measurement transformations. Furthermore, we show that this approach can significantly improve the parameter inference based on semi-quantitative data in comparison to available methods. Availability and implementation Modelers can easily apply our method by using our implementation in the open-source Python Parameter EStimation TOolbox (pyPESTO) available at https://github.com/ICB-DCM/pyPESTO.
BibTeX
@article{doresic2024efficient,
  author = {Domagoj Dorešić and Stephan Grein and Jan Hasenauer},
  title = {{Efficient parameter estimation for ODE models of cellular processes using semi-quantitative data}},
  journal = {Bioinformatics},
  year = {2024},
  doi = {10.1093/bioinformatics/btae210},
}
Figure