Human interpretation is a common practice in many scientific data analyses. After the data is processed to a certain extent, the remainder of the analyses is performed as a series of steps of processing and human interpretation. Many large scientific experiments span multiple organizations, therefore, both the data and the teams involved in these experiments, are distributed across these organizations. When the focus of an analysis is to extract new knowledge, collaboration is a key requirement. Real time or near real-time collaboration of expertise, on scientific data analyses, provides a better model of interpretation of the processed data. In this paper, we present a collaborative framework for scientific data analysis that is also secure and fault tolerant.