A pilot study for fragment identification using 2D NMR and deep learning.

This paper presents a proof of concept of a method to identify substructures in 2D NMR spectra of mixtures using a bespoke image-based convolutional neural network application. This is done using HSQC and HMBC spectra separately and in combination. The application can reliably detect substructures in pure compounds, using a simple network. Results indicate that it can work for mixtures when trained on pure compounds only. HMBC data and the combination of HMBC and HSQC show better results than HSQC alone in this pilot study.

Variable Data Independent Acquisition and Data Mining Exploring Feature-Based Molecular Networking Analysis for Untargeted Screening of Synthetic Cannabinoids in Oral Fluid.

Novel psychoactive substances (NPS) are constantly emerging in the drug market, and synthetic cannabinoids (SCs) are included in this NPS family. Forensic laboratories often struggle with these continually emerging SCs, forcing them to develop an untargeted workflow to incorporate these psychoactive drugs in their procedures. Usually, forensic laboratories select analytical methods based on targeted mass spectrometry (MS) technologies for strictly tracking already known NPS. The appropriate way to tackle unknown substances is to develop pipelines for untargeted analysis that include LC-HRMS analytical methods and data analysis. Once established, this strategy would allow drug testing laboratories to be always one step ahead of the new trends concerning the "designer drugs" market. To address this challenge an untargeted workflow based on mass spectrometry data acquisition and data analysis was developed to detect SCs in oral fluid (OF) samples at a low concentration range. The samples were extracted by mixed-mode solid-phase extraction and analyzed by Liquid Chromatography - High-Resolution Mass Spectrometry (LC-HRMS). Tandem mass spectra (MS2) were recorded performing a variable isolation width across a mass range of all theoretical precursor ions (vDIA) after the chromatographic separation. After raw data processing with the MSDial software, the deconvoluted features were sent to GNPS for Feature-Based Molecular Networking (FBMN) construction for nontargeted data mining. The FBMN analysis created a unique integrated network for most of the SCs assessed in the OF at a low level (20 ng/mL). These results demonstrate the potential of an untargeted approach to detect different derivatives of SCs at trace levels for forensic applications.