Explainable Artificial Intelligence (xAI) for 5-HT_2A Receptor Binding Affinity of New Psychoactive Substances

Area of Science:

• Pharmacology and Toxicology
• Computational Chemistry
• Machine Learning in Drug Discovery

Background:

• New psychoactive substances (NPS) pose significant public health risks due to their unpredictable potency and psychoactive effects.
• The 5-hydroxytryptamine receptor 2A (5-HT_2A) is a key target for many hallucinogenic drugs, including NPS.
• Assessing receptor binding affinity (Ki) is vital for understanding drug interactions but traditional in vitro assays are resource-intensive.

Purpose of the Study:

• To develop and validate machine learning models for predicting the binding affinity of NPS to the 5-HT_2A receptor.
• To utilize computational methods to accelerate the investigation of novel psychoactive compounds.
• To provide insights into the structure-activity relationships of 5-HT_2A ligands.

Main Methods:

• Collected publicly available Ki data for 5-HT_2A ligands.
• Calculated molecular descriptors and fingerprints for known NPS.
• Trained five classification machine learning models and employed explainable AI (SHAP values, similarity maps) for interpretation.

Main Results:

• Machine learning models achieved high predictive performance, with precision up to 93% and recall up to 92%.
• Explainable AI methods provided interpretable insights into the model's predictions.
• The study demonstrated the suitability of computational approaches for predicting 5-HT_2A ligand binding affinities.

Conclusions:

• Machine learning models can accurately predict the binding affinities of potential 5-HT_2A ligands.
• This computational approach offers a resource-efficient alternative to traditional in vitro binding assays.
• The findings support the use of AI in rapidly assessing the pharmacological properties of emerging NPS.