Boolean network identification from perturbation time series data combining dynamics abstraction and logic...

Area of Science:

• Systems Biology
• Computational Biology
• Bioinformatics

Background:

• Boolean networks and logic models are crucial for understanding complex signal transduction pathways.
• Current methods for learning these models often rely on static data or limited time points, assuming early steady states.
• This limits their ability to capture dynamic system behaviors.

Purpose of the Study:

• To generalize existing Boolean network learning methods to incorporate time-series phosphoproteomics data.
• To develop a method that can discriminate between Boolean networks based on their transient dynamics.
• To improve the accuracy and scalability of logic model learning for biological signaling.

Main Methods:

• Developed a generalized learning procedure that utilizes discretized time-series phosphoproteomics data.
• Identified a necessary condition for Boolean network dynamics to be consistent with time-series data.
• Employed Answer Set Programming (ASP) and model-checking for a global learning algorithm.
• Computed an over-approximation of Boolean networks fitting experimental data.

Main Results:

• Achieved a true positive rate exceeding 78% in learning logic models for mammalian signaling networks.
• Demonstrated computational efficiency, providing optimal answers for a large case study within 7 minutes.
• Quantified significant gains in prediction precision compared to methods using static data.
• Successfully identified potentially erroneous time-points in experimental data based on learned models.

Conclusions:

• The developed method effectively learns logic models from time-series data, capturing transient dynamics crucial for signal transduction.
• This approach offers a more accurate and scalable alternative to static data-based learning methods.
• The ability to identify data inconsistencies provides a valuable tool for experimental validation and refinement.