Compound identification in gas chromatography/mass spectrometry-based metabolomics by blind source separation.

Metabolomics GC-MS samples involve high complexity data that must be effectively resolved to produce chemically meaningful results. Multivariate curve resolution-alternating least squares (MCR-ALS) is the most frequently reported technique for that purpose. More recently, independent component analysis (ICA) has been reported as an alternative to MCR. Those algorithms attempt to infer a model describing the observed data and, therefore, the least squares regression used in MCR assumes that the data is a linear combination of that model. However, due to the high complexity of real data, the construction of a model to describe optimally the observed data is a critical step and these algorithms should prevent the influence from outlier data. This study proves independent component regression (ICR) as an alternative for GC-MS compound identification. Both ICR and MCR though require least squares regression to correctly resolve the mixtures. In this paper, a novel orthogonal signal deconvolution (OSD) approach is introduced, which uses principal component analysis to determine the compound spectra. The study includes a compound identification comparison between the results by ICA-OSD, MCR-OSD, ICR and MCR-ALS using pure standards and human serum samples. Results shows that ICR may be used as an alternative to multivariate curve methods, as ICR efficiency is comparable to MCR-ALS. Also, the study demonstrates that the proposed OSD approach achieves greater spectral resolution accuracy than the traditional least squares approach when compounds elute under undue interference of biological matrices.