Reference-free compound identification using computational prediction of molecular properties and multi-dimensional spectrometric measurements: a fentanyl case study

Abstract

Mass spectrometry is used to identify chemicals to which humans are exposed, but it cannot directly determine molecular structures. Instead, structures are inferred by matching experimental spectra to libraries of spectra constructed from analyses of pure reference compounds. However, the chemical space of human exposures far exceeds the amount of experimental library spectra. Here, we evaluate a reference-free strategy for confident identification of unknown molecules. Using fentanyl as a case study, we created a suspect library of over 1 billion computationally predicted fentanyl analogs and predicted molecular properties through machine learning, molecular dynamics, and density functional theory. Multi-dimensional spectra from a blinded analysis of a mock fentanyl tablet were matched with the predicted library, yielding an average of three candidate structures per measured analog, with six exact identifications. This work emphasizes the promise of reference-free molecular measurements for assessing human exposure by merging computational predictions with high-dimensional measurements.