Use of Independent Component Analysis and clustering methods to find and identify relevant components in a matrix of UV-spectral data

In the framework of the elaboration of new pharmaceutical formulations, statistical methodologies can accelerate and automate the development and optimization of quantitative methods.

To fulfil this objective, design of experiment (DOE) are widely used. In this context, the same mixture of analytes is injected while LC operating conditions are assessed. This gives plenty of very different chromatograms that can be tedious and time-consuming to interpret.

Recently, the independent component analysis (ICA) has shown its usefulness to interpret chromatogram, i.e. to separate numerical signals from a matrix containing data provided by liquid chromatography system equipped with ultra violet diode array detector (LC-UV DAD). A matrix containing peaks corresponding to different analytes is then obtained. A brief summary of the ICA algorithm, applied to this problem, will be first given.

The aim of the current work is to show that an automated methodology can be used to match together ICA-identified peaks that correspond to the same analytes in different chromatograms. In this way, this task, attributed to analytical experts, can be quickened and easier. This methodology uses classical hierarchical agglomerative clustering with special dissimilarity measures between spectra.