High-resolution accurate-mass mass spectrometry (HR/AM MS) is capable of trace-level, molecular characterization of organic contaminants in the environment. However, comprehensive sample characterization is impeded by inefficiencies in the structural annotation of detected features based on HR/AM MS analysis. We have developed a holistic, non-targeted analysis workflow to address these challenges, which combines HR/AM MS, cheminformatics, and computational MS. The approach utilizes a generalized linear model (GLM) to rank structural annotation based on multiple computational MS tools (i.e., MetFrag, CSI:FingerID, CFM-ID and MAGMa) and compound meta data predictive of potential environmental occurrence (i.e., literature and patent references). The model was trained on a set of 1355 HR/AM MS<sup>2 </sup>spectra from environmentally relevant compounds (e.g., industrial chemicals, pharmaceuticals and pesticides) and subsequently tested against a set of 1445 HR/AM MS<sup>2</sup>spectra generated from compounds not present in the training set. The trained GLM-based approach returned the correct structure annotation as the top-rank for &gt;75% of the test set, which was a significantly higher correct annotation rate than the computational MS tools implemented alone. Furthermore, the GLM-based approach yielded 141 top results that were not correctly identified by any of the computational MS tools. We subsequently applied the developed approach for the non-targeted structure annotation of chemical mixtures formulated as part of the EPA ENTACT trial. Mixtures were analyzed by LC-ESI(+)-HR/AM MS using an Orbitrap Fusion Lumos tribrid tandem MS. After feature detection, componentization, and alignment, high-performance distributed computing was utilized to perform automated molecular formula assignment, putative structure assignment and ranking based on the above-described GLM approach. Structure assignments by the GLM scoring approach were compared to conventional library scoring techniques and revealed that GLM scoring and spectral library searching returned the same top-ranked candidate for 409 of 496 (82%) library matches. This result demonstrates that GLM scoring has great potential for informing structure annotation efforts, even without extant library spectra. Overall, our results highlight the benefit of combining of state-of-the-art computational mass spectrometry tools with chemoinformatic approaches for increased efficiency and annotation rate in non-targeted analysis approaches.

Improving non-target identification of organic contaminants by probabilistic ranking of putative structure assignments by HR/AM MS(/MS) and computational mass spectrometry

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.