Premium
Automated QSAR with a Hierarchy of Global and Local Models
Author(s) -
Wood David J.,
Buttar David,
Cumming John G.,
Davis Andrew M.,
Norinder Ulf,
Rodgers Sarah L.
Publication year - 2011
Publication title -
molecular informatics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.481
H-Index - 68
eISSN - 1868-1751
pISSN - 1868-1743
DOI - 10.1002/minf.201100107
Subject(s) - quantitative structure–activity relationship , hierarchy , selection (genetic algorithm) , feature selection , computer science , machine learning , artificial intelligence , drug discovery , data mining , bioinformatics , biology , economics , market economy
We present an automated QSAR procedure that is used in AstraZeneca’s AutoQSAR system. The approach involves automatically selecting the most predictive models from pools of both global and local models. The effectiveness of this QSAR modelling strategy is demonstrated with a retrospective study that uses a diverse selection of 9 early stage AstraZeneca drug discovery projects and 3 physicochemical endpoints: Log D ; solubility and human plasma protein binding. We show that the strategy makes a statistically significant improvement to the accuracy of predictions when compared to an updating global strategy, and that the systematic biases inherent in the global model predictions are almost completely removed. This improvement is attributed to the model selection aspect of the strategy.