TransIA: A transformer-based few-shot learning framework for detecting PDE5 inhibitor derivatives

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Food Science

Background:

• Illicit additives (IAs), particularly phosphodiesterase-5 inhibitor (PDE5I) derivatives, are increasingly found in food and environmental samples.
• Conventional detection methods struggle to identify these IAs due to their designed evasion strategies.
• There is a critical need for rapid and accurate screening tools for food safety and environmental monitoring.

Purpose of the Study:

• To develop a cheminformatics tool for the rapid and accurate identification of PDE5Is from raw mass spectra.
• To address the growing health concerns associated with illicit PDE5I derivatives in various matrices.
• To improve food safety and environmental monitoring capabilities.

Main Methods:

• Development of TransIA (Transformer-based Illicit Additives Detection Network), a deep learning model utilizing transformer architecture.
• Training and testing the model on raw mass spectra data.
• Validation across diverse environmental and biological matrices.
• Deployment on real-world samples for IA identification.

Main Results:

• TransIA achieved 98% accuracy on the testing set.
• The model demonstrated a false-positive rate below 3%.
• Analysis of 197 real-world samples identified 21 IAs, including 16 previously unreported or uncatalogued by the model.
• Identified IAs included novel compounds and those not present in existing databases.

Conclusions:

• TransIA provides a highly accurate and efficient method for detecting illicit PDE5I derivatives.
• The developed approach significantly enhances the ability to identify novel and previously unknown illicit additives.
• This technology holds potential for integration with portable mass spectrometry for on-site environmental analysis and improved public health protection.