Towards an evolvable cancer treatment simulator

Area of Science:

• Computational biology
• Cancer research
• Bioinformatics

Background:

• High-fidelity computational simulations are crucial for advancing cancer therapy optimization.
• Increased simulation realism demands significant computational resources, posing a challenge for high-throughput analysis.
• Optimizing targeted drug delivery requires navigating complex biophysical parameter spaces.

Purpose of the Study:

• To investigate the efficacy of surrogate-assisted evolutionary algorithms for optimizing cancer therapy.
• To evaluate Gaussian process models and multi-layer perceptron neural networks as surrogate models within evolutionary algorithms.
• To enhance the efficiency of high-throughput multicellular simulations for therapeutic design.

Main Methods:

• Utilized PhysiCell, a multicellular simulator, for agent-based simulations of therapeutic compound delivery.
• Implemented evolutionary algorithms coupled with Gaussian process and neural network surrogate models.
• Explored the parameter space of biophysical properties to minimize cancerous cells post-treatment.

Main Results:

• Evolutionary algorithms effectively explored biophysical parameter spaces to minimize cancerous cells.
• Surrogate-assisted algorithms outperformed standard evolutionary algorithms in search efficiency.
• Demonstrated improved performance within a limited computational budget.

Conclusions:

• Surrogate-assisted evolutionary algorithms offer an efficient approach for optimizing cancer therapies using high-throughput simulations.
• This study represents a novel application of efficient evolutionary algorithms for maximizing tumor regression.
• The findings pave the way for accelerated discovery of optimal therapeutic designs.