Explainable Analysis for New Psychoactive Substance Identification with Chemical Insights

Area of Science:

• Chemistry
• Computer Science
• Pharmacology
• Public Health

Background:

• New psychoactive substances (NPS) present a growing global health risk due to their rapid evolution and structural diversity.
• Conventional detection methods struggle to identify novel NPS variants effectively.
• Deep learning offers a promising avenue for proactive identification of emerging substances.

Purpose of the Study:

• To introduce NPS-EDR, an Explainable Deep Reasoning model for the identification of new psychoactive substances.
• To develop a model that provides interpretable structural and functional analyses of potential NPS molecules.
• To enhance accuracy, precision, and transparency in NPS identification compared to existing methods.

Main Methods:

• Development of NPS-EDR, a two-stage prediction-explanation deep learning framework.
• Utilizing cooperative training of mode-specific experts and reinforcement learning for consistent predictions and explanations.
• Training on a chain-of-thought dataset of over 2,900 NPS and drug molecules, integrating chemical prior knowledge.
• Leveraging biochemical insights for structural-functional interpretation.

Main Results:

• NPS-EDR achieves superior accuracy, precision, and transparency in identifying potential NPS molecules.
• The model demonstrates enhanced analytical capabilities and builds confidence through its explanation framework.
• NPS-EDR outperforms mainstream large language models and biomolecular-specific chemical language models in NPS identification.
• The model provides interpretable structural and functional analyses, going beyond simple detection.

Conclusions:

• NPS-EDR offers a novel and effective approach to address the challenge of emerging new psychoactive substances.
• The model's transparent reasoning capabilities can significantly benefit public health strategies, pharmacological research, and forensic science.
• Explainable deep learning provides a powerful tool for proactive identification and analysis in the field of drug discovery and control.