Chemometric analysis of gas chromatographic peaks measured with a microsensor array: Methodology and performance assessment

Abstract

A hybrid multivariate curve resolution method that combines evolving factor analysis (EFA) with alternating least squares (ALS) is applied to simulated partially overlapping binary gas chromatographic (GC) peaks from a microsensor array detector. Extended disjoint principal component regression is then used to relate the results of EFA–ALS to vapor recognition probabilities. The application of this methodology to such data is illustrated and the performance is evaluated. Responses to a set of organic vapors obtained from a portable GC with a detector consisting of an array of four nanoparticle-coated chemiresistors (CR) are used to derive the absolute and relative sensitivity values for the modeling and simulations performed. From these, seven vapor pairs spanning a range of pattern similarity are selected and modeled as Gaussian peaks whose magnitudes and degrees of overlap are varied by simulation. Performance is assessed as a function of the response pattern similarity, chromatographic resolution, signal-to-noise ratio, and the relative response ratio of the composite peak constituents. Overall, despite the low dimensionality of the array data, EFA–ALS provides an effective means of extracting information about co-eluting components from the GC-microsensor array system, and the array provides sufficient diversity of responses to identify those components in most cases, provided that the relative response ratio is <20:1.