High-Precision and High-Throughput Screening in Environmental Samples Enabled by Secondary Mass Spectrometry Prediction: A Case Study on PPCPs.

Abstract

Machine learning (ML) has been extensively studied and developed for mass spectrometry prediction in recent years. However, a research gap exists regarding the integration of secondary mass spectrometry (MS2) prediction models with suspect screening for environmental samples analysis. To address this, we developed a high-throughput, high-precision (HTHP) screening strategy that integrates MS2 prediction with suspect screening for environmental sample analysis. The cosine similarity of the proposed model reached 0.8. Taking pharmaceuticals and personal care products (PPCPs) as the research object, we employed the message passing neural network (MPNN) architecture, a typical variant of graph neural network (GNN), to predict the MS2 information of compounds with unknown mass spectrometry profiles. In this study, ultra-high-performance liquid chromatography coupled with quadrupole-Orbitrap high-resolution mass spectrometry (UHPLC-Q-Orbitrap HRMS) was utilized for the screening in environmental samples. Concurrently, we established screening filtering criteria requiring a fragment match count of 19 or higher and a retention time deviation (|ΔRT|) value below 1.51 min. The HTHP strategy was applied to screen 38 target compounds and the results were compared with those obtained from targeted analysis achieving an overall identification accuracy of 87.5%, which is higher than the 78.6% accuracy of the suspect screening method. Ultimately, the HTHP screening strategy enabled the successful identification of 715 compounds, from which eight compounds were selected for confirmation using standard substances, successfully validating five compounds. This HTHP approach effectively enhanced the accuracy of suspect screening in environmental samples.

Abstract as supplied by the indexing source. © the authors / publisher.