RELATING BIOLOGICAL-ACTIVITY TO CHEMICAL-STRUCTURE USING NEURAL NETWORKS

In the last few years, neural networks have found increasing use in chemical applications, including their use in the analysis of quantitative structure-activity relationships (QSAR) data. Networks are able to perform the equivalent of discriminant and regression analyses, in addition to providing a novel method for the low-dimensional display of multivariate data. Experiments in our laboratories, using artificially structured data sets and real literature QSAR data with neural networks performing multiple linear regression, demonstrated their susceptibility to over-fitting, resulting in poor predicted abilities. Other network algorithms and training regimes are emerging in the literature which address these particular problems.