An Entropy-Based Approach to Model Selection with Application to Single-Cell Time-Stamped Snapshot Data.

Area of Science:

• Biophysics
• Computational Biology
• Systems Biology

Background:

• Single-cell experiments generate time-stamped snapshot (TSS) data, revealing cell-to-cell variability in protein abundances.
• TSS data offer insights into the statistical time-evolution of protein levels and cellular signaling kinetics.
• Model selection is challenging when multiple mechanistic models explain the same TSS data, especially when likelihood functions are unavailable.

Purpose of the Study:

• To develop and validate an entropy-based approach for selecting the most accurate mechanistic model from TSS data.
• To address the limitations of existing model selection methods that require likelihood functions.
• To provide a robust framework for analyzing single-cell protein dynamics and inferring underlying biochemical mechanisms.

Main Methods:

• An entropy-based approach utilizing split-sample techniques to leverage large datasets.
• Estimation of model parameters using generalized method of moments (GMM) software.
• Estimation of candidate models via kernel density estimators and a Gaussian copula.

Main Results:

• The proposed method successfully selected the correct "ground truth" model from competing mechanistic models using simulated data.
• Demonstrated the feasibility of using GMM software and kernel density estimators for model estimation in this context.
• Validated the approach's ability to assess relative model support through model selection probabilities computed via bootstrapping.

Conclusions:

• The developed entropy-based method provides a viable solution for model selection with TSS data, even without explicit likelihood functions.
• This approach enhances the analysis of single-cell protein dynamics and aids in understanding cellular signaling mechanisms.
• The study offers a computationally efficient and statistically sound framework for advancing quantitative systems biology.