BLOG 2.0: a software system for character-based species classification with DNA Barcode sequences. What it does, how to use it

In this paper we investigate logic classification and related feature selection algorithms for large biomedical data sets. When the data is in binary/logic form, the feature selection problem can be formulated as a Set Covering problem of very large dimensions, whose solution is computationally challenging. We propose an alternative approximated formulation for feature selection that results in an extension of Set Covering of compact size, and use the logic classifier Lsquare to test its performances on two well-known data sets. An ad hoc metaheuristic of the GRASP type is used to solve efficiently the feature selection problem. A simple and effective method to convert rational data into logic data by interval mapping is also described. The computational results obtained are promising and the use of logic models, that can be easily understood and integrated with other domain knowledge, is one of the major strengths of this approach.