UC Davis

The high-throughput ability and sensitivity of mass spectrometry make it the most popular platform for a metabolomics study. Thus, identifying small molecules from mass spectra plays a central role in metabolomics. There are many computational techniques for mass spectra raw data processing, including feature detection, peak alignment, and mass spectral deconvolution. This dissertation focuses on the conversion between spectral and structural information. It is a spontaneous thought to identify compound structure from mass spectra by searching the query spectra against a reference library with similarity score. However, this approach is limited by the availability of reference spectra and standard compounds. To bridge the gap, different computation tools based on fragmentation trees are developed to help annotate spectra. The other thought is to generate in-silico spectra from the structural information. Several compound databases, such as PubChem, KEGG, HMDB and CHEBI can be the source of structural information. Machine learning and heuristic approaches are trained from spectral knowledge to generate in-silico spectra. However, all those approaches cannot predict beyond the known data. As a first-principles method, Quantum chemistry modeling is only based on the rules of quantum mechanics and can help us explore the unknown space of metabolites. However, the quantum chemical simulation on molecules over 600 Da is too expensive to get accurate results. Chemical ionization with methane reagent gas can help identify the molecular ion species and help the unknown metabolite identification.This dissertation describes computational studies using molecular dynamics on in-silico mass spectra of small molecule generation. Chapter 1 provides an overview of applications of quantum chemistry to generate in-silico mass spectra. Chapter 2 showcases the performance of quantum chemistry simulations on small molecules and probes the parameter space to find potential improvements to the existing method. The conformational flexibility effect is also explored and has no correlation with prediction accuracy. Chapter 3 describes a new workflow to include chemical derivation, especially silylation in the quantum chemistry calculation. Different compound classes including organic acid, alcohol, amide, amine and thiols are simulated and compared against experimental mass spectra. The molecular dynamics trajectories are also investigated to find missed fragmentations from rearrangements. Chapter 4 provides a new algorithm that introduces excited state calculation into the molecular dynamics prediction of mass spectra. The new algorithm can predict more fragmentation reactions that are missing in previous studies and as a result, the mass spectral similarity scores are increased, and simulation is more accurate. Chapter 4 also discusses the limitations of molecular dynamics simulation time and the lack of rearrangement reactions. Chapter 5 provides another routine from the experimental side to help unknown metabolites identification with methane chemical ionization and quadrupole-time of flight mass spectrometry.