Analyse-IQ: machine learning software

Introduction Analyze-IQ  Hazard-DB  References & Further Info

This project was funded by Enterprise Ireland under the following grants: “Hazard-IQ: Intelligent Quantification of Raman Spectral Data for Hazardous and Illicit Substances”  Enterprise Ireland Technology Development Grant (TD/03/212), Commercialisation Plus supplement, 2007. “Hazard-IQ: Intelligent Quantification of Raman Spectral Data for Hazardous and Illicit Substances”  Enterprise Ireland Technology Development Grant (TD/03/212), 2003-06.

Introduction:

Analyze-IQ is a machine-learning based software package to analyze data from spectroscopic techniques such as Raman spectroscopy. For Forensic applications it will simplify and automate the process of identifying illegal narcotics, hazardous materials, explosives, and/or trace evidence. The software learns to recognize the presence (and quantity) of a target analyte in complex mixtures, yielding more accurate results than conventional chemometric methods.  

Existing spectral analysis software packages that are used for identification use relatively basic spectrum-matching for comparing the spectrum from an unknown sample with those of known pure materials that are available in digital databases.  Such packages do not work well when the substance being analyzed is composed of a complex mixture of compounds (for example, a narcotic cut with diluents). Analyze-IQ uses patented machine-learning algorithms to overcome this problem, by developing dedicated models for high value analytes such as illegal narcotics.  

The Analyze-IQ team comprised of Dr. Alan Ryder, Chemistry Department,  Dr. Michael Madden, IT Department, Dr. Kenneth Hennessy, and Dr. Tom Howley. 

The research matters article on these projects can be downloaded from here.  
A short briefing document is available from here.  

Back to top

Analyze-IQ:

The Analyze-IQ software package is a spectral analysis tool that uses advanced machine learning (and includes some chemometric methods) algorithms for both classification (ID) and quantification tasks in industry and research.  The main goal behind the software is to tackle analysis problems in complex mixtures such as those encountered in law enforcement and healthcare. 

The prototype version includes two novel machine learning methods (patent submitted) that are significantly better than conventional chemometric methods (data to be published soon). 
The Analyze-IQ package is fully integrated with the database and so allows the user the freedom to study very large and complex datasets such as those encountered in PAT, forensic, and biomedical applications. 

Back to top

Hazard-DB:

Hazard-DB is a spectral database package that will be available as a standalone or with ~2000 Raman spectra.  Hazard-DB has been designed from the ground up to accommodate mixtures, unlike most other spectral databases which only contain pure materials.  The ability to incorporate mixtures and to index the components (by CAS number etc.) is important for many different applications where mixtures are always encountered.  The database is also fully integrated with the Analyze-IQ package so that data can be extracted from the database to build very robust identification models.  These robust identification models can then be used for example to provide more accurate identification of cocaine containing samples or the presence of solvents in water.

The screenshot above right shows the Hazard-DB GUI, with spectra listings on the left and the Raman spectrum on the right.   The screenshot right shows the display panel for the sample information, in this case a mixture.  When inputting the spectral data, one can incorporate a wide range of additional data (manufacturer, lot number, appearance, purity, IUPAC name, common name, etc.). For mixtures, all the relevant data can be included, so that you can track mixtures that use the same materials by lot numbers.

Back to top

References

1. The effect of principal component analysis on machine learning accuracy with high dimensional spectral data.  T. Howley, M.L. O’Connell, M. Madden, and A.G. Ryder,  Knowledge-Based Systems, 19(5), 363-370, (2006).
2. Machine Learning Methods for Quantitative Analysis of Raman Spectroscopy Data, M.G. Madden and A. G. Ryder, Proc SPIE vol. 4876, 1130-1139, (2003). [ Abstract]

Back to top

• Disclaimer
• Privacy
• Copyright
• Accessibility