Accurate Deconvolution of Perfectly Coeluting Analytes by Exploiting Differential Expression Across Samples

Significance of the Topic

Chromatographic coelution of analytes impedes accurate identification and quantification in complex mixtures. Perfectly overlapping peaks often lead to mixed spectra and unreliable results, affecting metabolomics, environmental monitoring, pharmaceutical quality control and other applications. Leveraging differences in analyte abundance across multiple related samples can enable robust deconvolution even when retention times coincide exactly.

Study Objectives and Overview

This work presents an algorithmic approach to deconvolve perfectly coeluting compounds by exploiting their differential expression patterns across a sample set. Unlike traditional PARAFAC2-based methods that require manual factor estimation, this method automatically determines the number of components needed. The algorithm was validated on both synthetic GC-TOFMS simulations derived from NIST spectral libraries and on real GC-TOFMS metabolomics data.

Methodology

The data are structured as a three-way tensor of spectra, retention times and samples. Key steps include:

• Conversion of raw intensities to absolute ion counts followed by square-root transformation to stabilize variance.
• Folding spectra into chromatograms while unfolding chromatographic profiles across samples to avoid imposing strict trilinearity.
• Iterative factor analysis to decompose the tensor into spectral and concentration profiles.
• Automatic determination of factor number by removing features until residual variance aligns with Poisson ion statistics at the 3σ level.

Instrumentation Used

• Gas chromatography time-of-flight mass spectrometry (GC-TOFMS) for data acquisition.
• MATLAB R2014b (version 8.4.0.150421) for algorithm development and testing.

Main Results and Discussion

On synthetic data, including perfect coelutions of sec-butylbenzene and isobutylbenzene with p-cymene and simulated dead coelutions of TMS derivatives, the algorithm successfully recovered pure component spectra and concentration profiles. When analyte covariance across 15 samples was below r=0.9, both components were individually resolved with top-rank NIST library matches and high forward similarity scores (930–946 for oxalate derivative, 769–949 for furoate derivative). In real GC-TOFMS metabolomics data, near-perfect coelutions such as ethyl dodecanoate with siloxane and ethyl lactate with ethyl 2-hexenoate were accurately deconvolved, demonstrating applicability to practical workflows.

Benefits and Practical Applications

The algorithm reduces user intervention by auto-estimating component number, improves confidence in compound identification and quantification, and extends deconvolution capability to cases of complete peak overlap. These advantages support applications in metabolomics profiling, environmental analysis, process monitoring and quality assurance where coelution frequently hinders data interpretation.

Future Trends and Possibilities

Potential developments include comparative benchmarking against other deconvolution techniques, extension to multidimensional separations (GCxGC-MS, MS/MS), integration into vendor software platforms for real-time analysis, and application to large-scale clinical and industrial datasets where sample variability can be harnessed for improved resolution.

Conclusion

By exploiting differential expression across samples and using an automated factor selection strategy, the presented algorithm reliably deconvolves perfectly coeluting analytes without extensive user input. Validation on both synthetic and real GC-TOFMS data confirms its effectiveness and broad utility in analytical chemistry workflows.

References

• [1] Johnsen LG et al. Journal of Chromatography A 1503 (2017) 57–64
• [2] Humston-Fulmer EM et al. Metabolomics 2015 Proceedings, Poster 344