Data Interpretation by some Common Chemometrics Methods

This is part of an invited presentation to the Australian Research Committee workshop on Electrochemically Based Microsensing Arrays. Its objective was to demonstrate the use of the more common chemometrics methods for data analysis, especially pattern recognition. The substance of this article was to address the question of why is it advantageous to use chemometrics for multivariate data analysis? And if one decides to venture into this field, how one can approach this challenge. We address the ‘why’ question through a number of examples, which highlight some advantages of pattern recognition data analysis, particularly for very similar or complicated multivariate measurements such as spectra or cyclic voltammograms. The chemometrics method considered is the exploratory principal component analysis(PCA) and the associated data pretreament, PC and loadings plots as well as biplots; data classification models are mentioned as possible useful extensions beyond the pattern recognition stage. The examples are not necessarily based on electrochemical problems but the applications to electrochemical biosensors and associated measurements can be readily appreciated. The last of the four examples provided, briefly describes some prediction studies involving differential pulse polarographic measurements of mixtures, which produce overlapping polarograms. The chemometrics methods employed involve multiple linear regression(MLR), principal component regression(PCR), partial least squares(PLS) and the Kalman filter(KF) approach. Commonly available chemometrics software packages are briefly mentioned.