Deep denoising autoencoder-assisted continuous scoring of peak quality in high-resolution LC−MS data

Abstract

Accurate peak picking is a challenging but fundamental problem in LC-MS-based omics analysis. Previous efforts mainly focused on continuous wavelet transform (CWT) based peak detection, which is highly sensitive while suffers from unsatisfactory precision. Recently proposed deep learning (DL) based peak classifiers improve the performance significantly. However, their classification strategy loses the continuous criterion for controlling the false positive rate flexibly. Here we put forward AutoMS, which employs a deep learning-based denoising autoencoder to grasp the common characteristics of chromatographic peaks, and predict noise-deducted peaks from the original peak profiles. By comparing the difference before and after processed, it scores the peak quality continuously and precisely. From the evaluating result, AutoMS improved the accuracy for peak picking. AutoMS integrates HPIC for ROI extraction in order to accept raw data directly and output quantitative results. It also supports peak lists obtained from other tools with little adjustment. AutoMS is open source and available at https://github.com/hcji/AutoMS.