Figure 1

Fuzzy decisions tree and network. Quantum attributes are used to represent molecular systems. These attributes define a specific, relevant metric in the modelling structure space and thus allow the proper use of rigorous mathematical techniques. Also, they have strong relationships with chemical features controlling molecular interactions, and well-defined physical properties are often related to a single attribute. In its simplest implementation, the modelling algorithm produces a fuzzy decision tree, which can be generalized to a more powerful fuzzy decision network. Each tree node corresponds to a single attribute (interaction constraint). The attribute explaining most data variance occupies the highest node. In a fully resolved decision tree, terminal nodes contain only either active or inactive molecules. Further, each terminal node is fully characterized statistically by associated confidence intervals and other parameters. A decision network provides a complete characterization of the interaction patterns found within the modelling data.