Graph-Grammatiken zur Suche und Klassifikation von molekularen Strukturen
Date issued
Authors
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
License
Abstract
We define a graph rewriting system that is easily understandable by humans, but rich enough to allow very general queries to molecule databases. It is based on the substitution of a single node in a nodeand edge-labeled graph by an arbitrary graph, explicitly assigning new endpoints to the edges incident to the replaced node. For these graph rewriting systems, we are interested in the subgraph-matching problem. We show that the problem is NP-complete, even on graphs that are stars. As a positive result, we give an algorithm which is polynomial if both rules and the query graph have bounded degree and bounded cut size. We demonstrate that molecular graphs of practically relevant molecules in drug discovery conform with this property. The algorithm is not a fixed-parameter algorithm. Indeed, we show that the problem is W[1]-hard on trees with the degree as the parameter. Different approaches for learning graph grammars are presented. It is shown to what extent they can be used for substructure search or classification of molecules. Different aspects of production are considered, e.g., that a special structure is used, which was recognized in presented molecules. The results of the presented approaches are compared and evaluated with different machine learning approaches.