Using Emulation to Engineer and Understand Simulations of Biological Systems

Area of Science:

• Computational Biology
• Systems Biology
• Machine Learning in Biology

Background:

• Sophisticated biological simulators capture complex systems but pose challenges for statistical analysis.
• Intractable analyses due to time and resource limitations hinder the translation of simulation predictions into understanding.

Purpose of the Study:

• To apply emulation techniques for efficient evaluation and enhanced understanding of biological simulators.
• To demonstrate the utility of machine learning ensembles in accelerating complex simulation analyses.

Main Methods:

• Utilized surrogate models (emulators) to rapidly and accurately predict simulator outputs.
• Applied an ensemble of machine learning techniques to a lymphoid tissue organogenesis simulator.
• Extended the open-source statistical package 'spartan' with tools for emulator development and application.

Main Results:

• Machine learning ensembles reproduced complex statistical analysis results within seconds.
• Enabled previously intractable analyses, including multi-objective optimization and Approximate Bayesian Computation.
• Facilitated enriched evaluation and refinement of simulation insights.

Conclusions:

• Emulation significantly reduces computational requirements for analyzing complex biological models.
• Accelerated analyses lead to deeper biological understanding and improved model refinement.
• The enhanced 'spartan' package supports broader adoption of emulation in simulation-focused research.