Data-driven reverse engineering of signaling pathways using ensembles of dynamic models.

Area of Science:

• Systems Biology
• Computational Biology
• Bioinformatics

Background:

• Inferring dynamic signaling networks from experimental data is challenging due to system nonlinearity, limited measurements, and technological constraints.
• Lack of identifiability is a common issue in dynamic biological network modeling.
• Predicting responses to novel perturbations requires robust dynamic models.

Purpose of the Study:

• To present a novel methodology, SELDOM (enSEmbLe of Dynamic lOgic-based Models), for inferring and analyzing dynamic signaling networks.
• To develop a data-driven approach that does not require prior system knowledge for network inference.
• To enable reliable dynamic predictions for novel experimental conditions.

Main Methods:

• SELDOM builds an ensemble of logic-based dynamic models trained on experimental data.
• Mutual information is used to infer interaction networks as scaffolds for dynamic models.
• A model reduction step is included to prune spurious interactions and mitigate overfitting.
• Ensemble predictions combine individual model simulations for improved accuracy.

Main Results:

• SELDOM demonstrates highly competitive performance in recovering network topology compared to state-of-the-art methods.
• The methodology successfully builds dynamic models (based on ordinary differential equations) for mechanistic interpretation.
• SELDOM's ensemble predictions consistently outperform individual models and often surpass existing methods in predicting responses to new conditions.

Conclusions:

• SELDOM offers a robust framework for data-driven inference of dynamic signaling networks.
• The method enhances predictive capabilities for biological systems, moving beyond static network reconstruction.
• SELDOM provides a valuable tool for mechanistic understanding and prediction in systems biology research.