SLAW: A Scalable and Self-Optimizing Processing Workflow for Untargeted LC-MS

Metabolomics has been shown to be promising for diverse applications in basic, applied, and clinical research. These applications often require large-scale data, and while the technology to perform such experiments exists, downstream analysis remains challenging. Different tools exist in a variety of ecosystems, but they often do not scale to large data and are not integrated into a single coherent workflow. Moreover, the outcome of processing is very sensitive to a multitude of algorithmic parameters. Hence, parameter optimization is not only critical but also challenging. We present SLAW, a scalable and yet easy-to-use workflow for processing untargeted LC-MS data in metabolomics and lipidomics. The capabilities of SLAW include (1) state-of-the-art peak-picking algorithms, (2) a new automated parameter optimization routine, (3) an efficient sample alignment procedure, (4) gap filling by data recursion, and (5) the extraction of consolidated MS2 and an isotopic pattern across all samples. Importantly, both the workflow and the parameter optimization were designed for robust analysis of untargeted studies with thousands of individual LC-MSn runs. We compared SLAW to two state-of-the-art workflows based on openMS and XCMS. SLAW was able to detect and align more reproducible features in all data sets considered. SLAW scaled well, and its analysis of a data set with 2500 LC-MS files consumed 40% less memory and was 6 times faster than that using the XCMS-based workflow. SLAW also extracted 2-fold more isotopic patterns and MS2 spectra, which in 60% of the cases led to positive matches against a spectral library.