engleski

Artificial intelligence in ion chromatography

Operational costs of typical ion chromatographic (IC) analyses have increased considerably over the last years, although the nominal price of analysis has remained relatively stable. These higher operational costs need to be countered by the laboratories through better productivity of analytical equipment. This task might be, at least partially, accomplished by the approaches that emerge at the interface between IC method development and chemometrics. In particular, artificial intelligence as the most powerful chemometric tool has to be considered. This lecture presents a selection of the relevant issues that were recently coped with by our group. It will discuss applications of methodologies such as artificial neural networks (ANN), genetic algorithms (GA) and fuzzy logic (FL) in IC as well as some of their hybrids like neuro-fuzzy and neuro-genetic paradigms. Isocratic retention can be easily modeled using both simple regression and ANN methodology ; ANNs give only slightly better predictions, which does not completely justify their application due to their inherent complexity. However, when discussing gradient retention modeling, two issues are met. Firstly, the direct application of ANNs for gradient retention predictions is rather limited, i.e. ANN can be used only on a predefined domain of gradient profiles. Secondly, when using modeling based on transfer of data form isocratic to gradient elution (iso-to-grad) mode by the integration of gradient elution equation, the application of ANN seems to be essential. The core of the procedure is the isocratic model ; its slightly better description using ANN (with respect to simple regression modeling) will result in a significantly better prediction upon transfer to the gradient elution mode. This is in particular evident when using the iso-to-grad approach for peak shape prediction. The iso-to- grad approach can be made “experiment free” (for demonstration purposes only) by incorporating the data from the ion chromatographic column certificate. The powerful IC gradient signal prediction routines need an additional tool for pushing obtained information into practice, i.e. an optimization tool capable of finding optimal conditions in a virtually unlimited domain of chromatographic parameters (infinite number of possible gradient profiles). The most powerful optimization strategy is based on genetic algorithms. This results in a development of hybrid systems: neuro-genetic systems. Neuro- genetic systems, if properly designed, can successfully solve multi-criteria decision- making problems where, besides common ones, parameters such as ruggedness of the IC method can be incorporated. Nevertheless, the above mentioned methodology requires experimenting before modeling. Here the quantitative structure retention relationship (QSRR) modeling may be considered a reasonable step for further reduction of experimental effort and costs. GA-based selection of QSSR-descriptors exerted a significantly better performance than other methodologies such as principal component analysis combined with regression method. Further improvements are expected by replacing GA- selection with fuzzy logic selection of descriptors producing hybrid neuro-fuzzy systems. For its part, artificial intelligence attempts to cope with the ongoing challenges and to develop new tools to deal with new problems. Moreover, there are old chromatographic problems that could not be solved efficiently so far ; due to the increasing power of computers and due to progress in computer- related fields of human knowledge one may expect to solve these problems as well. However, a misuse of artificial intelligence might occur with those less familiar with data processing approaches and each application of artificial intelligence has to be taken cautiously.

artificial intelligence ; ion chromatography ; retention prediction ; method development

nije evidentirano