Smart computational exploration of stochastic gene regulatory network models using human-in-the-loop semi-supervised...

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Discrete stochastic models are crucial for understanding gene regulatory networks and predicting molecular interaction outputs.
• Model exploration, essential for hypothesis generation, is computationally demanding due to parameter uncertainty and requires extensive manual analysis of simulation results.
• Current methods limit systematic exploration to simpler models, hindering comprehensive analysis of complex biological systems.

Purpose of the Study:

• To develop an interactive and intelligent workflow for efficient exploration of gene regulatory network models.
• To reduce the computational burden and manual labor associated with analyzing large-scale simulation data.
• To enable modelers to rapidly identify ranges of interesting behaviors predicted by complex models.

Main Methods:

• Implemented a semi-supervised learning approach combined with human-in-the-loop data labeling.
• Utilized a feature space to group similar simulation outputs, allowing focused user input.
• Developed an interactive workflow that guides modelers in identifying and labeling significant behaviors.

Main Results:

• The developed workflow significantly reduces the time and effort required for manual inspection of simulation results.
• Modelers can efficiently discover ranges of interesting behaviors by interactively labeling data.
• This approach accelerates the process from initial model development to generating testable predictions.

Conclusions:

• The interactive workflow streamlines the exploration of discrete stochastic gene regulatory network models.
• Semi-supervised learning and human-in-the-loop labeling offer a powerful strategy for computational model analysis.
• This method substantially reduces manual workload, enabling faster progress in biological pathway and network modeling projects.