Non-linear QSAR modeling by using multilayer perceptron feedforward neural networks trained by back-propagation

• Published in: Talanta (Oxford) Vol. 56; no. 1; pp. 79 - 90
• Main Authors: González-Arjona, D, López-Pérez, G, Gustavo González, A
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 04.01.2002; Oxford Elsevier
• Subjects: Analytical chemistry; Artificial neural networks; Chemistry; Exact sciences and technology; General, instrumentation; Pattern recognition; QSAR; QSAR; Artificial neural networks; Pattern recognition; Agonist; Data analysis; Pharmacological activity; Molecular structure; Linear regression; Learning; PLS regression; Backpropagation algorithm; Feedforward neural nets; Structure activity relation; Multilayer perceptrons; Chemometrics
• ISSN: 0039-9140; 1873-3573
• DOI: 10.1016/S0039-9140(01)00537-9

The use of multilayer perceptrons (MLP) feedforward neural networks trained by back-propagation (BP) for non-linear QSAR model building is presented and explained in detail through a case study. This method was compared with others often used in this field, such as multiple linear regression (MLR), partial least squares (PLS) and quadratic PLS (QPLS). The case study deals with a series of 18 alpha adrenoreceptors agonists belonging to three different classes (alpha-1, alpha-2 and alpha-1,2) according to their different pharmacological effects. Each of them is described by 15 chemical features (the X block). Six pharmacological responses were also measured for each one to build the matrix of biological responses (the Y block). The results obtained indicated a slightly better performance of MLP against the other procedures, when using the correlation coefficient of the observed versus predicted response plots as an indicator of the goodness of the fit.