FIDDLE: a deep learning method for chemical formulas prediction from tandem mass spectra

Yuhui Hong,Sujun Li,Yuzhen Ye,Haixu Tang

DOI: https://doi.org/10.1101/2024.11.25.625316

2024-11-28

Abstract:Molecular identification through tandem mass spectrometry is fundamental in metabolomics, with formula identification serving as the initial step in the process. However, current computation-based methods for formula identification face challenges, such as limited accuracy and extensive running times, with some methods unable to predict formulas for relatively large molecules. The limitations may impede high-throughput workflows and diminish overall research efficiency and success. To address these issues, we introduce FIDDLE ( ormula entification by eep arning using mass spectrometry), a novel deep learning-based method for formula identification. Our training and evaluation dataset comprises over 38,000 molecules and 1 million tandem mass spectra (MS/MS) acquired by using various Quadrupole time-of-flight (Q-TOF) and Orbitrap mass spectrometers. Comparative analyses demonstrate that FIDDLE accelerates formula identification by more than 10-fold and achieves top-1 and top-5 accuracies of 88.3% and 93.6%, respectively, surpassing state-of-the-art (SOTA) methods based on top-down (SIRIUS) and bottom-up (BUDDY) approaches by an average of over 10%. On external benchmarking metabolomics datasets with novel compounds, FIDDLE significantly outperforms the SOTA methods, achieving on average the top-3 accuracy of 72.3%. Furthermore, combining FIDDLE with existing methods such as BUDDY further improves performance, which achieves a higher top-3 accuracy of 79.0%.

Bioinformatics