Peak Evaluation for Liquid Chromatography-Mass Spectrometry Data Using Neural Network on Statistical Quality Metrics

Peak picking is a crucial step in the data analysis workflow of Liquid Chromatography-Mass Spectrometry (LCMS) in untargeted metabolomics. The process is challenging due to interference from background signals such as noise and matrix effects. To address this challenge, we first collected 5,447 ion peaks and manually scored them after alignment. Next, we calculated 12 quality metrics for each ion peak. Finally, we developed a Multilayer Perceptron (MLP) model to perform regression analysis, as well as binary and ternary classification, making it adaptable to various application scenarios. The results demonstrate that the MLP-based peak evaluation method is effective in accurately predicting peak quality. Researchers can quickly initiate their work by setting just one or two score thresholds based on their specific needs. This approach retains the flexibility to handle different datasets without the need to adjust thresholds for each peak quality metric, representing a step towards more flexible and efficient peak picking.