Supervised learning in the gene ontology Part I: A rough set framework

Prediction of gene function introduces a new learning problem where the decision classes associated with the objects (i.e., genes) are organized in a directed acyclic graph (DAG). Rough set theory, on the other hand, assumes that the classes are unrelated cannot handle this problem properly. To this end, we introduce a new rough set framework. The traditional decision system is extended into DAG decision system which can represent the DAG. From this system we develop several new operators, which can determine the known and the potential objects of a class and show how these sets can be combined with the usual rough set approximations. The properties of these operators are also investigated.