LoMoGraph (Local Models for Graph Classification and Regression)

Fabian Buchwald, Tobias Girschick, Madeleine Seeland and Stefan Kramer, TUM, Germany

Many existing large databases cannot be used directly to build local (Q)SAR models, because they contain diverse sets of noncongeneric structures. We present LoMoGraph, a method that first detects structural clusters (i.e., clusters of molecules sharing a common structural scaffold) automatically and then uses those clusters to build local (Q)SAR models. The algorithm combines clustering and classification or regression for making predictions on chemical structure data. A structural clustering procedure is applied as a preprocessing step, before a (local) model is learned for each relevant cluster. Instead of using only one global model (classical approach), we use weighted local models for predictions of query compounds dependent on cluster memberships. The approach is evaluated and compared against standard statistical (Q)SAR algorithms on various datasets. The results show that in the majority of cases the application of local models significantly improves the predictive power of the derived (Q)SAR models compared to the classical approach, to models that are induced by a fingerprint-based or a hierarchical clustering approach and to locally weighted learning.

An example of a cluster of LoMoGraph is shown as well as sample predictions.

Additionally, an application of LoMoGraph as a web service in the OpenTox framework using the workflow management system Taverna is presented.

(presenting author: Fabian Buchwald)