Tag Archives: ASMS2025

Here, we describe the improvements of the computational workflow for SIL analysis already in place in the previous version of MARS (https://mass-analytica.com/products/mars/) a software solution for untargeted LC-MS untargeted metabolomics analysis.

In particular, MARS allows the generation of database of partially and/or uniformly labelled species starting from a database of native metabolite. Native species can be easy edited and labelled with common isotopes (e.g. D, 13C, 15N, 34S, 37Cl). MS and MS/MS data of labelled species are automatically computed from those of native metabolites and stored in the database for identification purpose. In addition, 9 ready-to-use databases including native and labelled metabolites reported in SIL studies using a specific isotope tracers are available for download. The isotopic pattern clustering algorithm was also recently optimized to take into consideration instrument resolution and isotopes of any element. Native and labelled compounds identified as different ionization adducts and at different

level (MS or MS/MS) can be inspected together or separately for an easier investigation and for projection into the metabolic maps included in the package.

This work used automation to generate experiments and process mass spectrometry data in a database for application in metabolite identification workflows. A script was executed, which directed the system to monitor a designated folder for an instrument sample list and LCMS data files. The presence of these files triggered both the creation of experiments in a database and the processing of the data for metabolite identification. Filters were applied to simplify analysis of the peaks before user intervention. Data processing and analysis were performed using MassMetaSite 4.7 in the ONIRO 1.6.2 server with LCMS data from Agilent, Bruker, Sciex, Thermo, and Waters.

Following incubation and LCMS data acquisition, the automation described herein significantly reduced user involvement in the laborious tasks of this metabolite identification workflow. The sample list, which was automatically generated from a mass spectrometer, was parsed to obtain the information necessary to define the protocol and create experiments within the ONIRO database. Using MassMetaSite, the LCMS data was analyzed to find metabolite peaks and elucidate the structures.

Several process tasks were also automated during data analysis to consolidate tasks which are otherwise tedious for the user during data review. Peaks were automatically filtered and discarded based on various parameters, such as mass error, isotope similarity score, and negative control area ratio. In experiments with multiple timepoints, calculations were performed to provide kinetic analysis, such as AUC and the area comparison between a given incubation sample and the 0 minute sample. An AI-based peak selection model was applied, which provided suggestions for peak selection and removal. The peaks which the model suggested to remove with high probability were automatically hidden. Finally, the system generated metabolite identification reports after approval of the experiments.

This workflow reduced the time required of a user by automating the parsing of a sample list, creation of experiments, processing of mass spectrometry data, and filtering of metabolite peaks. Upon review of the data by the user, the system automatically generated a metID report. This automation has the potential to significantly decrease the time between LCMS data acquisition and report generation, providing faster access to information to better understand the metabolism and design compounds with improved properties.

The methodology defines a new workflow that uses LC-MS data from peptide metabolic experiments as well as data coming from external sources to predict potential cleavage sites in new candidate’s peptide drugs by employing a machine-learning model. The models make use of transformer architecture with added mechanisms to encode graph structural data. Notably, these models eliminate the need for manual feature extraction, as they can predict peptide properties such as secondary structure and solvent accessibility. The methodology is designed to operate without structural constraints, allowing for linear and cyclic peptides, and including natural and unnatural amino acids. Users can train the models with their own experimental data. The methodology was validated using experimental MetID data from over 100 individual peptides.