Data processing tools comparison applied to clinical datasets for untargeted GC-MS-based metabolomics

This is the authors' abstract. We don't add key points for this paper.

Background In metabolomics studies, biological data are highly complex, requiring robust computational tools to detect biological variation in the organism under investigation. Despite this, comparative studies of data processing software for GC-MS data are scarce. This work evaluated the performance of two open-source software programs, XCMS and MS-DIAL, on three datasets: cells, serum, and tissue, acquired by GC-MS. Results Software performance was assessed based on the quality of the raw data matrix, the number of extracted features, the presence of missing values, and the number of annotated metabolites or chemical classes. Overall, both tools produced robust matrices and adequate statistical models, with high similarity in the regulation of common metabolites. For the investigated datasets, XCMS demonstrated slightly better performance in raw matrix quality, while MS-DIAL showed a higher correlation with the annotations. The combination of matrices broadened metabolic coverage, allowing for improvements in supervised statistical models. The choice of data processing platform and the parameters used directly impact the extracted data matrix and the conclusions regarding metabolic alterations. Significance These observations reinforce the need for careful evaluation in metabolomics analyses, so that selection involves not only parameters such as the number of features and metabolites obtained, but also the method's stability and its ability to show the chemical variability of the analyzed biological system.