Nested active learning for efficient model contextualization and parameterization: pathway to generating simulated...

Area of Science:

• Computational biology
• Systems biology
• Biomedical modeling

Background:

• Mechanism-based multi-scale computational models, like agent-based models (ABMs), are increasingly used for simulated clinical populations.
• Optimizing model context and content involves numerous free parameters, posing a significant computational challenge.

Purpose of the Study:

• To develop an efficient workflow for parameterizing and contextualizing an agent-based model (ABM) of systemic inflammation in sepsis.
• To significantly reduce the number of simulations required for model exploration.

Main Methods:

• Utilized a nested active learning (AL) workflow to optimize parameters for an ABM of systemic inflammation.
• Employed Artificial Neural Networks (ANNs) at two levels: mapping clinically relevant (CR) space and regressing CR space properties.
• Explored contextual parameters and internal model parameters related to signaling pathways.

Main Results:

• Reduced the number of simulations needed to map the CR parameter space by approximately 99%.
• Demonstrated the efficiency of the nested AL approach for complex models with numerous variables.
• Successfully parameterized and contextualized an ABM for sepsis simulation.

Conclusions:

• The nested active learning workflow offers substantial efficiency gains in computational modeling for biomedical research.
• This method accelerates the discovery and evaluation of diagnostic and therapeutic strategies for diseases like sepsis.
• The approach is scalable and likely to yield further improvements for more complex models.