Agent-based modelling approach to evaluate the effect of collaboration among scientists in scientific workflows

Automation in science is increasingly marked by the use of workflow systems (eg, Matlab) to facilitate the scientific discovery. The sharing of workflows through publication mechanisms supports the reproducibility and extensibility of computational experiments. However, the subsequent scientific discovery from a workflow relates to the level of collaboration among scientists. An agent-based model (ABM) is developed by coupling a scientific workflow with a model of scientist agents. The scientist agents are able to collaborate using a simplified small-world network. After a query is submitted to scientist agents, each scientist agent is able to extract data from data-sets, which are widely available online, using automated workflows to prepare a scientific report for a query. After data are collected from a workflow, data can be shared among scientists using one of the four collaboration scenarios, which simulate alternative level of data availability. Each scientist uses the data, which is collected from the database or through a shared environment, to deduce a scientific discovery. The ABM is demonstrated and evaluated for application within ecological science. Scientist agents collaborate and use the workflow tool, Kepler, to develop a linear regression model that captures the relationship between zooplankton populations and codfish population in the Norwegian Sea.