DuReS: An R package for denoising experimental tandem mass spectrometry-based metabolomics data

Abstract

Mass spectrometry-based untargeted metabolomics is a powerful technique for profiling small molecules in biological samples, yet accurate metabolite identification remains challenging. One of the primary obstacles in processing tandem mass spectrometry data is the prevalence of random noise peaks, which can result in false annotations and necessitate labor-intensive verification. A common method for removing noise from MS/MS spectra is intensity thresholding, where low-intensity peaks are discarded based on a user-defined cutoff or by analyzing the top "N" most intense peaks. However, determining an optimal threshold is often dataset-specific and may retain many noisy peaks. In this study, we hypothesize that true signal peaks consistently recur across replicate MS/MS spectra generated from the same precursor ion, unlike random noise. An optimal recurrence frequency of 0.12 (95% CI: 0.087-0.15) was derived using an open-source metabolomics dataset, which enhanced the dot product score between the experimental and library spectra by 66% post-denoising and resulted in a median signal and noise reduction of 5.83% and 99.07%, respectively. Validated across multiple metabolomics datasets, our denoising workflow significantly improved spectral matching metrics, leading to more accurate annotations and fewer false positives. Available freely as an R package, Denoising Using Replicate Spectra (DuReS) (https://github.com/BiosystemEngineeringLab-IITB/dures) is designed to remove noise while retaining diagnostically significant peaks efficiently. It accepts mzML files and feature lists from standard global untargeted metabolomics analysis software as input, enabling users to seamlessly integrate the denoising pipeline into their workflow without additional data manipulation.