Optimization of the ionization parameters in mass spectrometry does not guarantee a sufficient response and successful signal assignment when analyzing unknown compounds presenting signals of low intensity. The interpretation is difficult as the mass signals of degradation products are often less intense than background signals. Important information may be overlooked. Such critical and time consuming data analysis was overcome by developing a new strategy and open-source software called eicCluster offering unsupervised machine learning algorithms and powerful interactive visualization tools that made data processing fast and intuitive. Using eicCluster for high-performance thin-layer chromatography coupled with high-resolution mass spectrometry, mass signals of degradation products were highlighted in a stressed formulation, which were hardly found until linked to the new software. Owed to (1) preprocessing leading to intensity-agnostic signals and (2) the t-SNE algorithm, clustering mass signals based on their similarity, even compound ions present at low intensities were separated in subclusters from background signals (in silico highlighting). The resulting 2D maps allowed a new view on the data set to target molecules (degradation products) in complex mixtures. In addition to this new source of information, targeted on-surface synthesis of degradation products (in situ highlighting) was shown to support a fast structure elucidation when standards are not commercially available. It allowed a better understanding of the proposed degradation reactions in the formulation. As proof of principle, an original example formulation, its stressed samples as well as the proposed degradation products of on-surface synthesis were compared. In silico and in situ signal highlighting substantially eased data processing in structure elucidation.
Keywords: Data analysis; High-performance thin-layer chromatography; High-resolution mass spectrometry; On-surface synthesis.
Copyright © 2018 Elsevier B.V. All rights reserved.