Classification of High‐Activity Tiagabine Analogs by Binary QSAR Modeling

Andreas Jurik,Gerhard F Ecker,Barbara Zdrazil,Regina Reicherstorfer

doi:10.1002/minf.201300020

Abstract

Classification of High‐Activity Tiagabine Analogs by Binary QSAR Modeling

Highlights

In the present work, we describe a ligand-based approach to summarize SAR information derived from a dataset of published lipophilic aromatic GAT inhibitors
Indicator variables were introduced for the three scaffolds of the amino acid mimicry, namely R- and S-nipecotic acid and guvacine
The strategy was adjusted towards binary QSAR

Summary

Special Issue EuroQSAR

Andreas Jurik,[a] Regina Reicherstorfer,[a] Barbara Zdrazil,[a] and Gerhard F. In order to assess the quality of the models, internal validation by leave-one-out cross-validation and prediction of an external test set was performed For the latter, two procedures were applied to split the 162 compounds into 147 (90 %) for training and 16 (10 %) for testing. Introducing the indicator variables outlined above increased both positive and negative predictive power for the external test set from 42.9 % and 77.8 % to 60.0 % and 81.8 %, respectively, clearly justifying their use (Table 2). It were 9 mainly atom/bond count, adjacency matrix and polarity descriptors, performing well when compared to the VSA descriptors for the training set, but exhibiting inferior positive predictive power for the test set (Table 2). The best model showed an overall accuracy on the training set of 89.7 %, with 98.1 % for

Descriptor Description

Random splits

Findings

Computational Methods