A gray box framework that optimizes a white box logical model using a black box optimizer for simulating cellular...

Area of Science:

• Systems Biology
• Computational Biology
• Machine Learning

Background:

• Predicting cellular responses to perturbations is crucial for cell fate control but challenged by non-linear molecular interactions.
• Machine learning models offer potential for perturbation response prediction but often lack interpretability.
• Boolean networks are valuable for biological interpretation but optimizing large-scale networks is difficult.

Purpose of the Study:

• To develop a scalable and interpretable method for predicting cellular responses to perturbations.
• To address the challenges of non-linearity and interpretability in perturbation response prediction.
• To optimize Boolean network models for biological insights and predictive accuracy.

Main Methods:

• A scalable derivative-free optimizer trained by meta-reinforcement learning was developed.
• The optimizer was applied to Boolean network models for intracellular molecular regulation.
• The approach was tested for predicting anti-cancer drug responses in cancer cell lines.

Main Results:

• The optimized Boolean network model successfully predicted anti-cancer drug responses.
• The model provided interpretable insights into underlying molecular regulatory mechanisms.
• The meta-reinforcement learning approach proved effective for optimizing complex logical networks.

Conclusions:

• The novel optimizer enhances the predictive power and interpretability of Boolean network models.
• This approach facilitates reliable cell fate control through understanding molecular dynamics.
• The method offers a promising tool for systems biology and drug response prediction.