Optimization study with ligand-design interval rules

Issue title: MASOCR Workshop

Article type: Research Article

Authors: Paetz, Jürgen

Affiliations: Facbereich Biochemie, Chemid und Pharmazie, J.W. Goethe-Universität Frankfurt am Main, Institut für Organische Chemie und Chemische Biologie, Marie-Curie-Straβe 11, D-60439 Frankfurt am Main, Germany; Fachbereich Informatik und Mathematik, Institut für Informatik, J.W. Goethe-Universität Frankfurt am Main, Robert-Mayer-Straβe 11–15, D-60054 Frankfurt am Main, Germany. E-mail: [email protected]

Abstract: Drug design has emerged as an application area of soft computing methodology. To find potential novel drugs, up to millions of molecules need to be virtually screened by algorithmic techniques using computational devices. Due to the high combinatorial number of molecules experimental costs need to be decreased by improvements of the computational virtual screening method. In this contribution an adaptive neuro-fuzzy system is applied to find interval rules by cutting the adapted trapezoid membership functions, that provide knowledge about the important class of bioactive molecules as candidates for potential drugs. However, the aim is not to classify all the molecular data, but to find a small region, described by a rule, with a high enrichment of bioactive molecules. This small number of molecules could then be screened in the laboratory, limiting the costs clearly. The generated rules performed mostly better than common similarity measures and decision tree rules. Another part of the work herein was to increase again the perfomance of the best found rules so far by using evolutionary optimization. A comparison of four different parameter settings for the fitness function is given. The results of the hybrid approach are mainly more performant than the neuro-fuzzy interval rules solely.

Journal: Journal of Intelligent & Fuzzy Systems, vol. 17, no. 3, pp. 263-285, 2006