L.M.C. Buydens

RES, AN EXPERT SYSTEM FOR THE SET-UP AND INTERPRETATION OF A RUGGEDNESS TEST IN HPLC METHOD VALIDATION .1. THE RUGGEDNESS TEST IN HPLC METHOD VALIDATION

Abstract In method validation, an intralaboratory repeatability study and an interlaboratory reproducibility study can be performed as part of a precision test. In HPLC, an intralaboratory ruggedness test can be performed to detect problems that would otherwise be encountered in a reproducibility study. In a ruggedness test, variations in ambient factors that are expected to occur in practice, are simulated. Several steps determine the success of a ruggedness test. The complexity and lack of standard procedures for some of these steps is the main reason why ruggedness testing is still not widely accepted.

RES, an expert system for the set-up and interpretation of a ruggedness test in HPLC method validation: Part 2: The ruggedness expert system

Abstract An expert system is described for the design and diagnosis of the results of a ruggedness test in HPLC method validation. It advises on the selection of relevant factors and an appropriate experimental design and it translates the statistical results into chemically relevant results. If the test fails, the program can advise on how to improve the HPLC method. The basis of the program is a Common Data Structure which represents all objects that are important in HPLC. Linked to this Common Data Structure are modules that manipulate objects. The modules represent major steps in a ruggedness test. An expert system on factor choice and a program on solving resolution problems are integrated in this program as modules. The program is designed to advise on a complete ruggedness test, from factor choice to improvement of the method.

Lethalization, penalty and repair functions for constraint handling in the genetic algorithm methodology

A genetic algorithm was designed to find low energy distributions of ions above a crystal surface. To solve this constrained optimization problem, several constrained handling techniques were applied, i.e., lethalization, penalty and repair functions. It was shown that the simple lethalization scheme performs very well, and was at least comparable to some of the penalty functions. This was unexpected because, in general, it is believed that a severe penalization (i.e., lethalization) leads to poor results. An analysis of the fitnesses of trial distributions that violated a constraint as a function of the time, suggested that the properties of genetic based search were responsible for this result. From this it was concluded that a genetic algorithm in combination with lethalization may be an good choice to solve constrained optimization problems if the design of optimal penalty functions is difficult or impossible.

RES, an expert system for the set-up and interpretation of a ruggedness test in HPLC method validation: Part 3: The evaluation

Abstract Van Leeuwen, J.A., Buydens, L.M.C., Vandeginste, B.G.M., Kateman, G., Cleland, A., Mulholland, M., Jansen, C., Maris, F.A., Hoogkamer, P.H. and Van den Berg, J.H.M., 1991. RES, an expert system for the set-up and interpretation of a ruggedness test in HPLC method validation. Part 3: The evaluation. Chemometrics and Intelligent Laboratory Systems , 11: 161–174. In order to establish the confidence of target users in an expert system, it is important that it has been submitted to a thorough testing procedure. In the case of RES, an expert system on ruggedness testing in HPLC, the testing procedure is divided into two parts: a validation and an evaluation part. In the validation part, the performance of the expert system is measured against the experts performance using simulated test cases. In the evaluation part, the suitability of RES is assessed by evaluators in a real laboratory environment. RES has been evaluated in two laboratories. In one laboratory the emphasis of the evaluation was on the knowledge in RES. In the second evaluation the usability of RES with respect to users who were inexperienced in formalized testing was emphasized. The results and comments of the evaluators are discussed. RES has been found to be useful in practice.

The ineffectiveness of recombination in a genetic algorithm for the structure elucidation of a heptapeptide in torsion angle space. A comparison to simulated annealing

Abstract Genetic algorithms comprise a family of stochastic optimization strategies, which are often applied to solve complex optimization problems. The combination of population based search and a recombination operator distinguishes the genetic algorithm from other global optimization techniques that often only comprise a (sophisticated) mutation-selection scheme. Investigations were conducted that suggest, however, that recombination is not always effective, i.e., crossover was unable to recombine the so-called building blocks that should produce improved trial solutions. In this research the contribution of the crossover operation to the performance of the genetic algorithm was examined for a structure elucidation problem of a heptapeptide. In addition, the performance of the genetic algorithm was compared to the alternative simulated annealing strategy. It was shown that the current design of the genetic algorithm did not promote the recombination of building blocks, and was therefore easily outperformed by simulated annealing. The strategy presented to reveal the effectiveness of recombination is straightforward, and can easily applied to other genetic algorithm applications.

Expert systems in chemical analysis

Abstract The expert systems in chemical analysis (ESCA) project was set up to evaluate the merits of expert system technology for use in industrial chemical analyses. Its aim is to provide expert systems that will illustrate the benefits and shortcomings of expert system technology. The area of interest is high-performance liquid chromatography (HPLC) method development in pharmaceutical analysis. In the first stage a number of tools were tested with respect to their suitability for expert system building in HPLC method development. Expert systems were then built for specific subdomains of HPLC method development. The subdomains together cover the entire method development process. In the final stage, some of the systems will be integrated to form larger systems that provide strategies for consistent and efficient method development.

Learning classification rules from an ion chromatography database using a genetic based classifier system

Abstract A classifier system based on genetic algorithm methodology was developed for the automatic extraction of production rules from a database of about 6000 ion chromatography (IC) method examples. This machine learning strategy generated heuristics that can assist in the choice for a detection method for a specified set of IC method and solute properties. It was shown that the final set of rules proposed detectors that agreed with the database for 76% of the cases. Application to a separate test set showed a prediction ability of 82%. The database, because of the characteristics of the included cases, did not allow for a significant improvement of these results. However, the results are of significance for the further development of knowledge systems, which assist in the design of IC methods. Furthermore, this dataset comprised a considerable challenge to the applied machine learning method.

EXPERT SYSTEMS IN CHROMATOGRAPHY - RESULTS OF THE ESCA PROJECT

Abstract The final results of the ESCA project (Expert Systems for Chemical Analysis) are presented. This is one of the major projects in the field of expert systems for chromatography. Expert systems have been developed that cover the important areas of method development in LC. In the last part of the project attention was concentrated on two issues, the study of integration possibilities of the different stand-alone systems and the important aspect of validation and evaluation of the developed expert systems. The integration studies and the results of the validation and evaluation are discussed.

VALIDATION AND REFINEMENT OF EXPERT SYSTEMS - INTERPRETATION OF NMR-SPECTRA AS AN APPLICATION IN ANALYTICAL-CHEMISTRY

Abstract Wehrens, R., Buydens, L. and Kateman, G., 1991. Validation and refinement of expert systems: interpretation of NMR spectra as an application in analytical chemistry. Chemometrics and Intelligent Laboratory Systems , 12: 57–67. In this article we propose a strategy for the validation and improvement of expert systems in analytical chemistry. An application of this approach in two-dimensional NMR spectrum interpretation is shown, and the results indicate that significant performance improvements can he achieved in this domain.

Expert systems for multivariate calibration, trendsetters for the wide-spread use of chemometrics

Abstract Gerritsen, M., Van Leeuwen, J.A., Vandeginste, B.G.M., Buydens, L. and Kateman, G., 1992. Expert systems for multivariate calibration, trendsetters for the wide-spread use of chemometrics, Chemometrics and Intelligent Laboratory Systems , 15:171–184. Chemometrics is less applied in practice than desirable. One of the reasons is the lack of software tools that allow an easy application of chemometrics by non-experts. A possible solution to this problem is the integration of expert systems with software for data analysis. This is illustrated with an example on high-performance liquid chromatography—ultraviolet data analysis. The integration of expert knowledge with chemometrical software will probably become an important future trend in chemometrical research as it is not restricted to multivariate data analysis.