An Evaluation of Methods for Inferring Boolean Networks from Time-Series Data

Area of Science:

• Bioinformatics and computational biology
• Systems biology
• Molecular biology

Background:

• Regulatory networks govern cellular behavior and decision-making.
• Boolean networks are a common computational model for gene regulation, where gene states are binary (on/off).
• Inferring these networks involves data discretization and subsequent network learning.

Purpose of the Study:

• To evaluate combinations of data discretization and Boolean network learning methods.
• To assess the performance of these combinations on diverse regulatory systems.
• To provide open-source implementations of the studied methods.

Main Methods:

• Considered three data discretization methods, including a novel approach.
• Evaluated three Boolean network learning algorithms.
• Tested all nine possible combinations on four biological regulatory systems with varying complexity.

Main Results:

• The choice of discretization and learning methods significantly impacts Boolean network accuracy and predictive power.
• Contrary to some views, Boolean networks can accurately model biological dynamics.
• Sufficient time points in experimental data are crucial for effective network inference.

Conclusions:

• Optimized combinations of discretization and learning methods yield robust Boolean models of gene regulatory networks.
• This work challenges previous assessments of Boolean network capabilities.
• Publicly available open-source implementations are provided to facilitate further research.