Applicability Domain Estimation

Machine learning algorithms infer a predictive model by generalizing the relationship between features of the molecules and target properties that are part of the training set. Thus, a key assumption is that the learned relationship applies not only for the structures from the training set, but also for other molecules and therefore can be exploited to estimate the (unkown) target property of these compounds. Unfortunatey, due to the high structural diversity of molecules, it cannot be expected for machine learning based models, which rely on a specific training set, to give reliable results for all possible compounds. Thus, it is important to consider the subset of the chemical space, in which the model is applicable. The approaches to this problem that have been proposed so far mostly use vectorial descriptor representations to define this domain of applicability of the model, which cannot be extended easily to structured kernel-based machine learning models. For this reason, several approaches to estimate the domain of applicability of a kernel-based QSAR model have been developed in the department (Fechner et al., J.Chem.Inf., 2010).

References

Estimation of the applicability domain of kernel-based machine learning models for virtual screening
Nikolas Fechner, Andreas Jahn, Georg Hinselmann and Andreas Zell
in Journal of Cheminformatics, 2:2
Abstract,Free Full Text PDF, DOI: 10.1186/1758-2946-2-2
Kernel-based estimation of the applicability domain of QSAR models
Nikolas Fechner, Georg Hinselmann, Andreas Jahn and Andreas Zell
in Journal of Cheminformatics, 2010, 2(Suppl 1), P38.
Abstract, PDF DOI: 10.1186/1758-2946-2-S1-P38
Estimating the applicability domain of kernel based QSPR models using classical descriptor vectors
Nikolas Fechner, Georg Hinselmann, Christina Schmiedl and Andreas Zell
in Chemistry Central Journal, 2008, 2(Suppl 1), P2.
Abstract, DOI: 10.1186/1752-153X-2-S1-P2

Contact

Nikolas Fechner, Tel.: (07071) 29-77174, nikolas.fechner@uni-tuebingen.de

Georg Hinselmann, Tel.: (07071) 29-77174, georg.hinselmann@uni-tuebingen.de

Andreas Jahn, Tel.: (07071) 29-77175, andreas.jahn@onlinehome.de