Supporting scientific collaboration through class-based object versioning

Reuse of scientific data is central to much of science. Although data produced by individual researchers and groups is made publicly available, effective sharing is often prevented by lack of common resource discovery mechanisms and by format interoperability issues. Unlike commercial databases that operate fixed programmes (e.g. mortgage plan) and variable data (e.g. interest), in a scientific environment the reverse applies and the methods to process the data changes while the original data items themselves stay unchanged. Scientists often build on existing work and try different techniques for processing datasets, necessitating changing methods. In this paper, we provide a class-based object versioning framework that supports dynamic changes to pipelines while managing dependencies. The framework addresses the management of arbitrary changes made to scripts during a data flow and the association of these changes to data created.