OdoriFy: A conglomerate of artificial intelligence-driven prediction engines for olfactory decoding.

Area of Science:

• Computational chemistry
• Cheminformatics
• Molecular biology

Background:

• Olfaction, the sense of smell, is complex and underexplored due to molecular intricacies.
• Odorant receptors (ORs) are crucial for detecting diverse odorants, initiating olfactory signaling.
• Limited computational tools and unknown odorant-receptor interactions hinder understanding of olfactory function.

Purpose of the Study:

• Introduce OdoriFy, a web server for predicting odorant-receptor interactions using deep neural networks.
• Provide tools for identifying odorants for specific ORs, classifying chemicals, finding responsive ORs, and validating interactions.
• Offer explainable AI insights into predictions at atomic and amino acid levels.

Main Methods:

• Development of OdoriFy, a web server integrating deep neural network prediction engines.
• Implementation of four key modules: Odor Finder, Odorant Predictor, OR Finder, and Odorant-OR Pair Analysis.
• Utilizing explainable AI for transparent prediction rationales and a curated dataset of human ORs, agonists, and nonagonists.

Main Results:

• OdoriFy successfully predicts odorant-receptor relationships, aiding in understanding olfactory mechanisms.
• The server provides predictions with rationales, enhancing interpretability.
• Comparative analysis showed OdoriFy's performance is comparable to structure-based ligand interaction methods.

Conclusions:

• OdoriFy is a valuable, freely accessible web resource for researchers studying olfaction.
• The tool facilitates the identification of odorant-receptor pairs and chemical classification.
• Explainable AI integration in OdoriFy enhances the understanding of molecular interactions in smell.