go to UNSW home page
UNSW logo School of Chemistry Home Page
> Contacts   > Sitemap
About the School divider Research divider Graduate Study divider Undergraduate Study divider High School Resources

Research Groups

Chemometrics And Metrology Group (CMG) - Research Interests

Chemometrics

Chemometrics is the application of mathematics and statistics in chemistry and related disciplines. During my academic career I have been involved with teaching, development and application of various chemometrics methods including linear and non-linear calibration, genetic algorithms, neural networks, experimental design, PCA and PLS and alike. In recent years however I am mostly involved with Bayesian statistical inference and two- and multi-way data analysis.

Statistical inference

Cover of book Data Analysis for Chemistry

Statistical inference allows us to make statements on the probability of finding data given an hypothesis (the so-called frequentists approach) or the probability of an hypothesis given the data (Bayesian approach). We work with the NSW Department of Environment and Climate Change on matching oil spills with putative sources, and the National Measurement Institute on the heroin signature program. We have developed a new Bayesian approach for matching spectra that has direct applications in forensic investigation of petroleum oil spills (see details below). I have developed and teach a postgraduate course (CHEM7115) on data analysis and experimental design in chemistry. With colleague Justin Gooding I have published the material I have been teaching in this course in a new introductory book 'Data Analysis for Chemistry.'  

Bayesian inference - some details

Bayesian probability propagation

Pr(H| E) is the probability of the hypothesis H given the evidence E ( for example the probability of an oil spill being matched with a possible source  given the correlation coefficient of IR spectra of the spill sample and suspect source sample). 

Pr(E | H) is called the likelihood of the evidence given the truth of the hypothesis, and can be obtained from the distribution of data from a library of pairs of samples of known status (matched = H is true, or not matched not H is true)

Pr(H) and Pr(not H) are the prior probabilities of the two hypotheses before any evidence is considered. See this publication by Jianfeng Li and Brynn Hibbert (PDF).

Multi-way Analysis

Since 2004 we have been investigating multi-way analysis and its applications in chemistry and biology. This field of research is quickly growing in the group and we are keen to develop this further. We are now working on exploring new aspects and applications of multi-way methods such as PARAFAC, PARAFAC2, GEMANOVA and N-PLS for various applications including environmental forensics, microbiology, electrochemistry and organometallic chemistry. For more details see Diako's webpage.