Comparing evolutionary optimization with ant colony optimization of drug design interval rules with and without pre-initialization

Many applications deal with knowledge in the form of ´if-then rules´. In numerical data spaces the condition part of such rules is often based on intervals where the values of single variables are allowed to be within the ranges of the intervals. The interval rules can be interpreted geometrically as hyper rectangles. They can be derived heuristically by adaptive learning. In a previous approach we used cuts of membership functions of neuro-fuzzy rules for the pre-initialization of interval rules, that were reduced in their dimensions. As we showed before, such rules can be optimized by evolutionary or by ant colony algorithms to a problem-specific criterion. We demonstrate how interval rules in the chemical area of virtual screening can be optimized to characterize molecules as novel drugs. Mainly, a comparison between evolutionary and ant colony optimization is given, with and without using the neuro-fuzzy pre-initialization of interval borders. The results show that pre-initialization is more useful for the evolutionary optimization paradigm