Automated Gas Chromatography/ Mass Spectral Decomposition and Analysis: Tools for Automating and Improving the Use of GC/MS Instruments

S.E. Stein, O. Toropov (Contractor), J. Klassen, W.G. Mallard, and J.J. Reed

Objective: To develop and test algorithms for automatically deconvoluting and analyzing GC/MS data files using a target library of compounds.

Problem: The program has been funded by the Defense Threat Reduction Agency (DTRA) to provide a method for analyzing for chemical weapons banned under the Chemical Weapons Convention. The software implementing the algorithms must provide full blinding of the analysis process not to compromise the proprietary data of treaty participants. In general, the manual analysis of GC/MS data files for complex mixtures can be time consuming and error prone. The normal method of doing a background subtraction to extract the single component can be essentially impossible in a complex mixture because there is no background. Even in only moderately complex chromatograms, a manual subtraction can produce seriously erroneous results. In addition, the use of retention information to reduce false positives is far more efficient with computer techniques.

Approach: A detailed noise analysis is performed, followed by a deconvolution of each of the peaks in the total-ion gas chromatogram. The resulting components are then compared to reference spectra using a series of algorithms to emulate the degree of confidence that an analyst would have in the deconvoluted peak. The process of extracting the distinct components out of a complex data file breaks down into four parts: noise perception and evaluation, component perception, signal extraction, and compound identification. The noise perception and evaluation step is central to the analysis because the recognition of the difference between a "real" peak caused by a compound eluting from the column and a "false" peak caused by noise depends upon a knowledge of the nature and size of the noise. Once the noise is understood, the individual components are extracted. The extraction of the signal involves examining the overlap of components and removing mass spectral peaks associated with a different component. Ongoing testing involves a number of laboratories both in the United States and abroad where specific chemical agent samples are examined.

Results and Future Plans: The algorithm was tested extensively by using a target library of chemical weapons agents. Over 40,000 data files were examined to ensure that the algorithm does not produce false positives. At the same time, a number of experiments was performed by other laboratories with low concentrations of chemical agents to demonstrate that the algorithm is sensitive enough to detect true positives at analytically useful concentrations. The results of these tests have shown that the algorithms used in the development of the software are robust and capable of automated and blinded analysis. This year, Version 2 of the software was released. This version includes a number of small changes in the algorithm that resulted from the testing effort as well as the inclusion of the ability to process a number of new instrument file formats.

The use of retention indices is central to the further reduction of false positives. The software developed here has been adapted by the Organization for the Prohibition of Chemical Weapons (OPCW) for use in all inspections involving GC/MS instrumentation. Work to improve the ability to predict retention index data for chemical agents from structural information and from physical property data on analogous compounds is ongoing.

Publications: