Premium
Radial Basis Function Neural Networks Based QSPR for the Prediction of log P
Author(s) -
Yao XiaoJun,
Liu ManCang,
Zhang XiaoYun,
Zhang RuiSheng,
Hu ZhiDe,
Fan BoTao
Publication year - 2002
Publication title -
chinese journal of chemistry
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.28
H-Index - 41
eISSN - 1614-7065
pISSN - 1001-604X
DOI - 10.1002/cjoc.20020200805
Subject(s) - quantitative structure–activity relationship , molecular descriptor , linear regression , artificial neural network , chemistry , stepwise regression , correlation coefficient , biological system , radial basis function , function (biology) , artificial intelligence , mathematics , statistics , stereochemistry , computer science , evolutionary biology , biology
Quantitative structure‐property relationship (QSPR) method is used to study the correlation models between the structures of a set of diverse organic compounds and their log P . Molecular descriptors calculated from structure alone are used to describe the molecular structures. A subset of the calculated descriptors, selected using forward stepwise regression, is used in the QSPR models development. Multiple linear regression (MLR) and radial basis function neural networks (RBFNNs) are utilized to construct the linear and non‐linear correlation model, respectively. The optimal QSPR model developed is based on a 7‐17‐1 RBFNNs architecture using seven calculated molecular descriptors. The root mean square errors in predictions for the training, predicting and overall data sets are 0.284, 0.327 and 0.291 log P units, respectively.