Gene regulatory networks with binary weights.

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Binary neural networks utilize 1-bit precision for weights and activations, reducing storage requirements.
• Gene regulatory network (GRN) modeling aims to understand complex biological interactions.
• Applying binary representations to GRNs could offer computational efficiencies.

Purpose of the Study:

• To develop an evolutionary computation framework for learning binary threshold networks.
• To adapt binary neural network principles for gene regulatory network modeling.
• To evaluate the framework's performance on established biological models.

Main Methods:

• An evolutionary computation framework was designed to infer binary threshold networks.
• Differential evolution and particle swarm optimization algorithms were employed.
• The framework was tested on GRN models for bacterial quorum sensing and yeast cell-cycle regulation.

Main Results:

• The framework successfully inferred binary threshold networks for both biological models.
• Models with strictly binary weights and thresholds achieved minimal error (2 bits).
• Relaxing the binary restriction for activation thresholds resulted in networks with zero error.

Conclusions:

• Binary threshold networks offer a viable and efficient approach for modeling gene regulatory networks.
• Evolutionary computation provides a robust method for inferring these binary networks.
• The findings suggest potential for simplified yet accurate GRN representations in computational biology.