AddictedChem

AddictedChem predicts and evaluates new psychoactive substances (NPS) by integrating DEA-controlled-substance data with machine learning models to support identification and analysis of addictive compounds.

Key Features:

• Comprehensive Database: Consolidates data from the US Drug Enforcement Administration (DEA) on controlled substances including physical and chemical characteristics, literature classified by MeSH, target binding data, pharmacokinetics (absorption, distribution, metabolism, excretion, and toxicity; ADMET), and related genes, pathways, and bioassays.
• Predictive Modeling: Implements 29 predictive models based on seven molecular descriptors and machine learning, with top models achieving BA 0.940 and AUC 0.986 on the test set and BA 0.919 and AUC 0.968 on an external validation set.
• Consensus Strategy: Applies a consensus strategy that integrates multiple predictive models to enhance reliability of NPS identification.
• Chemical Space Construction: Constructs a chemical space of vectorized addictive compounds to represent diversity and macro-level relationships among existing NPS.
• Machine Learning Algorithms: Employs five different machine learning algorithms to develop predictive models using molecular descriptors.
• Data Integration: Aggregates diverse datasets from authoritative sources including the US Drug Enforcement Administration (DEA).
• Validation: Assesses model performance via rigorous testing and external validation.

Scientific Applications:

• Rapid Identification: Enables swift identification of new psychoactive substances to support regulatory agencies and public health officials.
• Research Tool: Provides data and predictive outputs to study pharmacological properties and biological targets of controlled substances.
• Public Health and Safety: Facilitates early detection and evaluation of NPS to mitigate public health risks associated with these substances.

Methodology:

Aggregating DEA datasets; developing 29 predictive models using seven molecular descriptors and five different machine learning algorithms; applying a consensus strategy; constructing a chemical space of vectorized compounds; and validating models via testing and external validation.