OPTIMIS: Optimizing Personalized Therapies through Integrated Multiscale Intelligent Simulation

Area of Science:

• Computational Medicine
• Systems Biology
• Artificial Intelligence

Background:

• Controlling complex biological systems across scales is challenging due to molecular variability influencing macroscale behavior.
• Deterministic models neglect molecular variability, while stochastic simulations are too slow for AI training.

Purpose of the Study:

• To develop a novel AI framework for adaptive intervention in multiscale biological systems.
• To enable high-throughput simulations for training artificial intelligence in computational medicine.

Main Methods:

• Developed a hybrid AI framework combining Gillespie algorithm (microscale) and ordinary differential equations (macroscale).
• Created a differentiable Neural Ordinary Differential Equation (Neural ODE) surrogate for fast digital twin simulations.
• Utilized deep reinforcement learning (RL) agents to learn closed-loop treatment policies.

Main Results:

• The AI framework successfully learned dynamic, closed-loop treatment policies for engineered cellular therapy.
• Microscopic cellular activity was used as an early-warning signal to adjust drug dosage.
• Improved successful control rates to over 70% in highly unstable simulated phenotypes.

Conclusions:

• The AI framework provides a practical and generalizable approach for adaptive intervention in multiscale biological systems.
• This computational advance facilitates controlling complex diseases by integrating molecular variability.
• The developed framework accelerates AI training for biological system control.