Abstract
Targeted, suspect and non-targeted screening by high-resolution mass spectrometry (HRMS) is developing rapidly. In this study, a qualitative screening method was established using HPLC-HRMS on data dependent acquisition for the analysis of mycotoxins in maize. To ensure the sensitivity and applicability of the method, 41 mycotoxin standards were applied for method optimization. A quantitative structure-retention relationships (QSRR) model was developed for retention time prediction and projection using machine learning, providing supplementary evidence for molecule annotation. The predicted errors were all below 0.5 min, contributing to improve the confidence level of suspect and non-targeted screening for mycotoxins. Thresholds affecting the accuracy of screening results were also investigated systematically. Performance metrics including Accuracy, F1 score, Matthew's correlation coefficient (MCC) were introduced to evaluate the qualitative screening method. The developed method was applied in the qualitative screening of collected maize samples, where 11 mycotoxins were screened at high confidence level.