“RAMclustR/RAMsearch: A New Strategy for Compound Identification in Non-targeted Metabolite Profiling Studies”
Professor Jessica Prenni, Colorado State University, Fort Collins, Colorado
Mass spectrometry based non-targeted metabolite profiling is a powerful approach that involves an unbiased analysis of the metabolome, is hypothesis generating, and has the potential for novel discoveries. However, despite the growing interest in this approach, a significant bottleneck remains in the annotation of retention-time specific mass signals (features) which requires extensive manual (and therefore subjective) interpretation. Electrospray ionization (ESI) is widely utilized in LC-MS workflows and while ESI is a relatively soft ionization method, biological compounds frequently give rise to multiple mass signals due to complex in-source fragmentation, adduction and multimer formation. Thus, novel tools for feature grouping that accurately reflect individual metabolites are necessary for more efficient, accurate, and comprehensive annotation of the metabolome. To address this bottleneck we have developed RAMclustR, a clustering workflow based on both correlation and co-elution similarities which can accurately determine this feature relationship for every detectable metabolite. RAMclustR can be utilized for both traditional MS data as well as for data independent acquisitions. Further, we have developed a modeling approach that uses molecular descriptors and is trained on spectra from authentic standards, to predict in-source peak patterns and retention time for compounds that we do not have standards for (e.g. entire HMDB library of ~40K compounds). Metabolite annotation is then based on in-source spectral matching against spectral libraries (experimental or predicted) using our recently developed tool RAMsearch, allowing for more rapid, accurate, and complete annotation of non-targeted MS-based metabolomics datasets.