Feeding graph machine learning into the agent based model for network analysis

Agent-based modeling is a powerful computational tool that can simulate complex behavior based on rules in both micro and macro scales. Defining the constant agent behavior patterns rules is uncertain since it depends on many facts leading to complicated decision-making processes. Integrating machine learning, specifically into a complex multiscale multi-agent model, could provide scientists with practical tools to make decisions. It also allows him to execute unlimited datasets in a large parametric ABM simulation on HPC/ Cloud infrastructures to improve accuracy, accelerate time to solution and significantly reduce costs.

This paper presents a modeling and simulation systems workflow that allows us to mine network measures and visualizing the evolving behavior of a tumor agent based model at both molecular and cellular levels. This workflow is proposed to construct a complex, scalable network based on our simulated tumor growth agent-based model. Three applicable machine learning techniques are  combined with the traditional simulation to train the tumor network evolving. These techniques also discover and classify subgraphs.

In this Model, the front-end integration is the strength of Python modeling based on packages such as an ABM package (Mesa), analysis, and visualization packages (Numpy, SciPy, Matplotlib). On the other hand, the Tumor simulation is getting more orchestrated while united with machine learning packages (NetworkX) or distribution utility modules and third-parties (Pypi) and container facilities deep at the back-end. 

Link to the code : https://github.com/ghta1000/Mesa-NetworkX-Agent-model.git