F. Questier

Classification and regression tree analysis for molecular descriptor selection and retention prediction in chromatographic quantitative structure-retention relationship studies.

The use of the classification and regression tree (CART) methodology was studied in a quantitative structure-retention relationship (QSRR) context on a data set consisting of the retentions of 83 structurally diverse drugs on a Unisphere PBD column, using isocratic elutions at pH 11.7. The response (dependent variable) in the tree models consisted of the predicted rention factor (log kw) of the solutes, while a set of 266 molecular descriptors was used as explanatory variables in the tree building. Molecular descriptors related to the hydrophobicity (log P and Hy) and the size (TPC) of the molecules were selected out of these 266 descriptors in order to describe and predict retention. Besides the above mentioned, CART was also able to select hydrogen-bonding and molecular complexity descriptors. Since these variables are expected from QSRR knowledge, it demonstrates the potential of CART as a methodology to understand retention in chromatographic systems. The potential of CART to predict retention and thus occasionally to select an appropriate system for a given mixture was also evaluated. Reasonably good prediction, i.e. only 9% serious misclassification, was observed. Moreover, some of the misclassifications probably are inherent to the data set applied.

Determining orthogonal chromatographic systems prior to the development of methods to characterise impurities in drug substances.

To define starting conditions for the development of methods to separate impurities from the active substance and from each other in drugs with an unknown impurity profile, the parallel application of generic orthogonal chromatographic systems could be useful. The possibilities to define orthogonal chromatographic systems were examined by calculation of the correlation coefficients between retention factors k for a set of 68 drugs on 11 systems, by visual evaluation of the selectivity differences, by using principal component analysis, by drawing color maps and evaluating dendrograms. A zirconia-based stationary phase coated with a polybutadiene (PBD) polymer and three silica-based phases (base-deactivated, polar-embedded and monolithic) were used. Besides the stationary phase, the influence of pH and of organic modifier, on the selectivity of a system were evaluated. The dendrograms of hierarchical clusters were found good aids to assess orthogonality of chromatographic systems. The PBD-zirconia phase/methanol/pH 2.5 system is found most orthogonal towards several silica-based systems, e.g. a base-deactivated C16 -amide silica/methanol/pH 2.5 system. The orthogonality was validated using cross-validation, and two other validation sets, i.e. a set of non-ionizable solutes and a mixture of a drug and its impurities.

Exploratory chemometric analysis of the classification of pharmaceutical substances based on chromatographic data.

A chemometric study has been conducted on a published data set consisting of the retention times of 83 substances, from five pharmacological families, on eight HPLC systems. Principal component analysis, clustering and sequential projection pursuit were applied. In this way it was investigated to what extent the combination of chromatography and chemometrics allows one to make conclusions about pharmacological activities of (candidate) drugs and what the contribution is of the different HPLC systems considered.

Ruggedness testing of chromatographic methods: selection of factors and levels

The first step in a ruggedness test is the selection of factors to be examined and their levels. In this paper, both topics are discussed, thereby completing a strategy described earlier. It is demonstrated, by means of some examples, that depending on the formulation (definition) of a factor, information that is physically more or less meaningful is extracted from the experimental design results. Among others, the inclusion of the compounds of a buffer and of the components of a mixture in a screening design were examined. A general guideline to select the levels of the factors in a ruggedness test was proposed. Some special cases, i.e. asymmetric intervals around the nominal level, were also discussed.

Application of rough set theory to feature selection for unsupervised clustering

Abstract Feature selection is a valuable technique in data analysis for information-preserving data reduction. This paper describes the use of rough set theory (RST) to construct reducts in a supervised way for reducing the number of features in unsupervised clustering. The application to a hierarchical clustering of Pseudomonas species is presented as an example. The Wallace measure is used for the comparison of the clustering results based on the original data set and those based on the reduced data set.

A ruggedness test strategy for procedure related factors: experimental set-up and interpretation.

A strategy to perform ruggedness tests for mainly procedure related factors is described. The different steps in the set-up of the experiments and in the interpretation of the results are given. The described strategy is based on a number of case studies and allows a statistical interpretation of the significance of the effects. It was implemented in a software tool. This original strategy was completed with a number of minimal screening designs which reduce the number of experiments to perform, but in consequence only allow a limited or no statistical interpretation of the effects. Some of the minimal designs are expandable to designs with characteristics similar to those of the original strategy.

Data structures and data transformations for clustering chemical data

The quality of a clustering of chemical data is determined by a proper choice of distance measures and data transformations. The latter aspect is often neglected and its importance is shown here. It is also shown that the V-shaped data structure that is often obtained in a principal component analysis of chemical data may indicate that the clustering of the raw data can lead to classifications that are not relevant from a chemical point of view and that the log double centering transform should be considered as a possible alternative.

RTS, a computer program for the experimental set-up and interpretation of ruggedness tests

A computer program is described for the experimental set-up and interpretation of ruggedness tests. The implemented strategy was based on a number of case studies and contains both recommended designs and minimal designs. The minimal designs reduce the number of experiments, but they cannot be statistically interpreted based on the interaction or dummy factor effects. The use of randomization tests as an alternative statistical interpretation method for the significance of the effects was examined. Some of the minimal designs are expandable to designs with characteristics similar to those of the recommended designs. The program is designed to facilitate the selection of the designs and the interpretation of the results and to prevent or detect problems such as drifting of responses.

The Neural-Gas network for classifying analytical data

This article introduces the Neural-Gas network, a fast neural net-based method for clustering, and shows how it is applied to gas chromatographic patterns of Maillard reaction products. The advantages of Neural-Gas are compared to the K-means clustering method and one of the best-known neural methods for clustering, the Kohonen self-organising maps. Some novel combinations with visualization techniques are also presented.

Feature selection for hierarchical clustering

Abstract Feature selection is a valuable technique in data analysis for information-preserving data reduction. This paper describes a feature selection approach for hierarchical clustering based on genetic algorithms using a fitness function that tries to minimize the difference between the dissimilarity matrix of the original feature set and the one of the reduced feature sets. Clustering trees based on reduced feature sets are comparable with those based on the complete feature set. Special measures to favor small reduced feature sets are discussed.