AI-Powered Decoding of Microbial Omics Signatures in Multidimensional Mass Spectrometry

Omics molecules are produced and transformed in networks of biochemical reactions in microbial systems. Multidimensional mass spectrometry (MS) data, including liquid chromatography and ion mobility spectrometry separations, and collected in data-independent acquisition mode, contain rich information to better characterize and study proteins, metabolites, and lipids, in complex samples. Artificial intelligence (AI) has the potential to exploit this data more efficiently than conventional MS tools. We developed PeakDecoder2 leveraging an AI-based computer vision model. Raw data was converted to MZA and Python code was implemented to: 1) read a library of molecules containing expected retention time, collision cross-section, and fragmentation pattern, 2) extract signals from raw data and generate images, and 3) perform inference to score the presence of molecules. Training was performed with curated images and labels from proteomics and metabolomics data. The model is able to identify precursors and their fragments in new samples through automated detection of overlapping regions. For data collection, LC methods were optimized (C18 column with 30 min gradient for proteomics and lipidomics, and HILIC column with 7 min gradient for metabolomics) and coupled to an Agilent 6560 Drift Tube IM-QTOF-MS. Pseudomonas Putida was used as complex background and 290 isotopically labeled peptides were spiked at 10 different concentrations in triplicates. PeakDecoder2 performed fully automated processing of 30 runs and 580 targets (290 light and heavy peptides), scoring 17,400 instances in 17 hours on a desktop computer (in contrast to months and thousands of clicks using existing tools). The number of heavy peptides detected correlated with concentrations (30-261), and light peptides were detected consistently in all samples (83 on average). Work is in progress to improve performance and evaluate fungal and bacteria samples relevant in biotechnology and environmental research.