A chemical AI and data science platform for the identification and discovery of novel psychoactive substances

Princeton University, Princeton NJ

Investigators

Abstract

Project Abstract/Summary ––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––– Novel psychoactive substances (NPS), including synthetic cannabinoids, cathinones, benzodiazepines, and opioids, are increasingly a cause of intoxications and fatalities and a major public health concern. Toxicological laboratories are tasked with detecting NPS in clinical and forensic samples in order to diagnose intoxications, direct medical treatment, and provide epidemiological data that informs public health responses. However, the chemical diversity of NPS, and the rapid pace at which NPS appear on the illicit market, makes this a daunting task. Because individual laboratories cannot realistically develop clinical assays for every NPS, there are gaps even in the detection of known NPS. Meanwhile, the detection of unknown NPS remains a highly time- and labor-intensive process that is performed primarily within highly specialized laboratories. To address these issues, this project will develop a comprehensive NPS database and innovative artificial intelligence (AI) tools to assist toxicological laboratories in identifying both known and unknown NPS. In Aim 1, we will assemble and curate the most comprehensive data resource of known NPS in existence. In Aim 2, we will repurpose neural network architectures that power widely-used language models to predict the structures of as-of-yet unknown NPS likely to emerge on the illicit market in the future. Because mass spectrometry is the primary analytical technique used to detect NPS within toxicological laboratories, we will then produce data resources that facilitate the identification of both known and unknown NPS by mass spectrometry. Specifically, we will apply a suite of computational and machine-learning tools to predict the mass spectrometric properties (e.g., MS/MS and retention time) of both the known and unknown NPS that would be observed if these compounds were present in a human tissue or biofluid. These data resources will be disseminated via an interactive web application that is already widely used in the toxicological community, and via in silico databases that can be integrated directly within existing NPS detection workflows by toxicologists without any computational training. By enabling faster and more comprehensive NPS detection, these resources will enhance clinical diagnoses, improve toxicological testing, and better equip public health systems to respond to emerging drug threats.