Artificial neural networks in the optimization of microemulsion liquid chromatography retention and resolution
Book part (Published version)
MetadataShow full item record
Artificial neural networks (ANNs) present a powerful tool for the modeling of chromatographic retention and resolution as they provide the ability to predict the effects of changes in chosen chromatographic parameters under various conditions. The tracking of certain chromatographic parameters presents the most important part in the optimization of any chromatographic method. One common situation is that in which opposing objectives (e.g., minimization of run time and maximization of chromatographic resolution) must be achieved. For this reason, different kinds of eluents are used in liquid chromatography (LC) methods, offering different mechanisms of retention and separation. The choice of eluent(s) mainly depends on the chemical structures and physical and chemical properties of the analytes. In addition, depending on the researchers' experience and knowledge, different optimization costs are acceptable. In this chapter, the feasibility of ANNs for modeling a chromatographic system w...as evaluated using the example of a microemulsion LC (MELC) separation of perindopril tert-butylamine and its structurally similar impurities (perindoprilat, Y31, Y32 and Y33). The selected complex mixture and finally defined microemulsion eluent composition demonstrated the application of ANNs in LC method optimization in a most descriptive and comprehensive way. In accordance with the situation and previous experience, a multiple-layer perceptron (MLP) was selected; MLP is the ANN form most commonly used for retention modeling. For the evaluation of the ANNs, a set of thirty experiments defined by central composite design (CCD) was performed. The influence of five factors on the chromatographic system was examined. ANNs were used in anticipation of the retention behavior of the eluting substances as well as of the chromatographic resolution (nine outputs in total). To obtain the networks with best characteristics, two separate networks (the first for retention factors and the second for resolution factors) were constructed from the experimental results. In this way, the number of nodes in the input and output layers were defined so that the network topologies for retention and resolution factors were 5-x-5 and 5-x-4, respectively. By employing D-optimal design in the first part of network optimization, the number of nodes in the hidden layer and the number of experimental data points used for training were simultaneously investigated. Furthermore, a series of training algorithms was applied to the current MLP network. The Backpropagation (BP), Conjugate Gradient-descent (CG), Quick Propagation (QP), Quasi-Newton (QN), and Delta-bar-Delta (DBD) algorithms were used to obtain the optimal network. The predictive ability of the optimized neural network was evaluated using several statistical tests. Consequently, the successful application of ANNs in the optimization of an LC method in a pharmaceutical analysis was demonstrated via the body of analyzed and discussed data.
Source:Artificial Neural Networks, 2011, 387-410
- Nova Science Publishers, Inc.