2010 IR Workshop

Frank Vogt

[Click here for the Wednesday talk abstract]

[Click here for the Thursday talk abstract]


Vogt received his degree in physics from the Technical University of Karlsruhe/Germany in 2000. In his graduate research he developed a novel spectroscopic technique for environmental analytical chemistry. After graduation he spent one year in a governmental lab in Germany working on remote spectroscopic imaging. His research focus switched to analytical chemistry when he received a postdoctoral scholarship from the German Research Foundation to join Boris Mizaikoff at Georgia Tech (2001-2003) followed by a PostDoc with Karl Booksh at Arizona State (2003-2005).

At both positions Vogt developed chemometric algorithms (multivariate statistics in chemistry) for quantitative analyses of key compounds in environmental samples. In 2005, he accepted a faculty position at the University of Tennessee where he is establishing a research group combining spectroscopic imaging and chemometrics of with tasks.

Vogt ’s research interests center around data analysis with applications in analytical chemistry including signal statistics, image analysis, multi-component quantifications in biomaterials as well as pattern recognition for sample classification. His group conducts environmental analytical chemistry utilizing spectroscopic imaging on heterogeneous samples such as microalgae cells and develops advanced chemometric software for quantitative analyses of chemically complex biomaterials as well as for classification of samples obtained under ill-conditions experimental conditions.


Spectrochemical Sample Characterizations  – a Primer in Chemometrics
This presentation introduces data analysis methods for two types of sample characterization – quantification of key analytes and classification/discrimination of samples based on their infrared-spectroscopic signature. For the first topic, Beer’s Law and (univariate) calibration curves will be briefly reviewed and their limitations for real-world applications outlined.

These concepts are then expanded in a very intuitive way towards Classical Least Squares (CLS) which can perform multi-component quantification of analytes with overlapping spectra. In typical real-world applications, however, calibration information is only available for a few key compounds which are often embedded into a largely unknown chemical matrix; furthermore, measurement conditions and/or samples may not be stable long enough and thus can introduce additional sources of variance.

In such measurement tasks, CLS fails for the lack of sufficient information about the samples and resulting violations of the experimental assumptions required for the CLS algorithm; nonetheless, CLS opens the door to understand more advanced techniques. The aforementioned challenges are handled by algorithms like Principal Component Regression and Partial Least Squares (PLS) which require a more mathematical approach but achieve reliable concentration predications with fewer experimental conditions/requirements.

The second topic in this presentation focuses on sample discrimination (same/different) by comparing their spectroscopic signatures. A first, straightforward approach is based on a correlation analysis of two spectra which measures the similarity among them independent from concentration or pathlength. It is only a sample step from discrimination to classification as in the latter case, one of the spectra was acquired from the unknown and the second from a known calibration sample.

This discussion will be visualized with examples from spectroscopic experiments performed on E. coli and microalgae cells. However, biomaterials are known for limited sample reproducibility and thus unknown features are often found in the spectra reducing the correlation and thus misguiding a classification. In order to incorporate ‘allowed’ sample variation, this correlation was generalized to compare an unknown to a number of class representatives.

This presentation will guide the audience through the most important steps of basic chemometrics and will be supported by a detailed tutorial which will be available prior to the talk. However, due to the number of chemometric methods only an introduction of the most important concepts can be given.

Evaluating High-fidelity Spectroscopic Image cubes – Planned Chemometrics at SRC
Based on the previous presentation, applications of chemometrics developed in the Vogt laboratory to spectroscopic studies of microalgae cells performed by Hirschmugl at the SRC will be discussed. Two research directions are underway, i.e., quantification of key compounds found in algae such as proteins, amino acids, carbohydrates, DNA, lipids, carboxylic acids and algae species identification and studies of chemical impacts imposed by environmental conditions.

These two topics will serve as introduction of a remote-accessible ‘virtual chemometrics lab’ which is currently under development. As research in chemometrics, the compilation of spectroscopic calibration databases, and imaging spectrometers are not required to be located in the same location, a broad range of researchers can potentially take advantage of a remotely accessible data analysis facility. Into the ongoing software design it is factored in that the typical user would only have detailed knowledge about their experiments but is not required to be familiar with the code details. 
This presentation is intended to lead into an open discussion about applications of spectroscopic imaging+chemometrics and required software tools.