Development and in silico evaluation of large-scale metabolite identification methods using functional group detection for metabolomics

Abstract

Large-scale identification of metabolites is key to elucidating and modeling metabolism at the systems level. Advances in metabolomics technologies, particularly ultra-high resolution mass spectrometry (MS) enable comprehensive and rapid analysis of metabolites. However, a significant barrier to meaningful data interpretation is the identification of a wide range of metabolites including unknowns and the determination of their role(s) in various metabolic networks. Chemoselective (CS) probes to tag metabolite functional groups combined with high mass accuracy provide additional structural constraints for metabolite identification and quantification. We have developed a novel algorithm, Chemically Aware Substructure Search (CASS) that efficiently detects functional groups within existing metabolite databases, allowing for combined molecular formula and functional group (from CS tagging) queries to aid in metabolite identification without a priori knowledge. Analysis of the isomeric compounds in both Human Metabolome Database (HMDB) and KEGG Ligand demonstrated a high percentage of isomeric molecular formulae (43 and 28%, respectively), indicating the necessity for techniques such as CS-tagging. Furthermore, these two databases have only moderate overlap in molecular formulae. Thus, it is prudent to use multiple databases in metabolite assignment, since each major metabolite database represents different portions of metabolism within the biosphere. In silico analysis of various CS-tagging strategies under different conditions for adduct formation demonstrate that combined FT-MS derived molecular formulae and CS-tagging can uniquely identify up to 71% of KEGG and 37% of the combined KEGG/HMDB database vs. 41 and 17%, respectively without adduct formation. This difference between database isomer disambiguation highlights the strength of CS-tagging for non-lipid metabolite identification. However, unique identification of complex lipids still needs additional information.

Perspectives and expectations in structural bioinformatics of metalloproteins

“…Through the use of the scientific method, research moves forward in a generally self‐correcting fashion as scientists evaluate the claims, methods, and results of each other. We have publications that pointed out issues in published analyses, published software, and public databases . However, we have tried to raise these issues with an open and collegial tone, with due diligence in fully checking other authors' results, and, in certain cases, contacting the authors in order to fully understand and check our own criticisms.…”

Inferring metabolite interactomes via molecular structure informed Bayesian graphical model selection with an application to coronary artery disease

“…To generate informative shrinkage priors for the adaptive Bayesian graphical Lasso, we utilized a local structure similarity metric. This metric was adapted from the previously described Chemically Aware Substructure Search (CASS) algorithm (Mitchell, Fan, Lane, & Moseley, 2014). In this adaptation, the structural similarity between any two chemical structures ( and ) was estimated using strings representing local chemical structure (referred to as the atom's color) centered at every atom in the two structures.…”

Using fragmentation trees and mass spectral trees for identifying unknown compounds in metabolomics

“…Recently, CASS ( C hemically A ware Sub-structure S earch) was developed to provide a tool that automatically detects functional groups in compound libraries [81]. CASS is also designed to create a functional group-resolved metabolite database.…”