Characterization of chemical “dark matter” is critical for human metabolome and exposure assessment, but the identity of many biologically-active compounds will remain unknown until analytical chemistry approaches are developed that (i) can identify compounds without reliance on data from analysis of authentic reference materials, (ii) have resolving power sufficient to resolve chemically similar compounds and (iii) have methods for comparing experimental data to reference libraries of chemical properties while accounting for the inherent accuracies in aspects of all associated analytical and computational pipelines. To begin addressing this need, our team has developed a suite of tools that collectively assembles evidence for the presence of small molecules in complex samples through the use of libraries containing multiple measured and calculated chemical properties. Here, we discuss the advancement of these tools and a current evaluation of this suite of standards-free identification methods on a set of natural products and microbial toxins. Furthermore, we evaluated four existing and new methods for quantifying chemical space: (i) principal components calculated from over 50 chemical properties, (ii) K-means clustering on these properties, (iii) distribution of ClassyFire superclasses, and (iv) relative location in deep learning latent space. The results show that compounds with similar behavior (e.g. detectable via MS) are best grouped using clustering, rather than by overall chemical class. Finally, we discuss how these methods may enable determination of a conservative false discovery rate for molecular identification and lead to generation of novel candidate molecules that do not correspond to any currently known molecules.

<i>In silico</i> metabolite property libraries and quantitative chemical space analysis: Path toward novel molecule identification and false discovery assessment

2019 Fall National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.

In silico metabolite property libraries and quantitative chemical space analysis: Path toward novel molecule identification and false discovery assessment