Machine learning–driven mapping of the odorant–olfactory receptor–olfactory perception network: A comprehensive review

Olfaction plays an essential role in human survival, evolution, and well-being. Mapping the odorant – olfactory receptor (OR) – olfactory perception network is essential for understanding olfactory mechanisms and identifying odorants, and machine learning approaches can facilitate this process. Odor quality and threshold are important olfactory perception characteristics for various industries. Moreover, ORs are promising biosensor materials for odor measurement. This review focuses on mapping the odorant – OR – olfactory perception network using machine learning methods. We comprehensively summarize and categorize the latest odor-related data sources for molecular properties, ORs, and olfactory perceptions. The current prediction models and general prediction workflows are reviewed. We delve into the application of machine learning in exploring the relationships among odorants, ORs, and olfactory perception. Deep learning methods based on molecular graphs (area under the receiver operating characteristic curve reaching 0.964) outperform classical machine learning approaches in odor quality prediction. Random Forest models (square correlation coefficient reaching 0.798) generally show advantages over other classical models in predicting odor threshold. We highlight the role of ORs in network mapping. Integrating machine learning and molecular docking can accelerate the identification and application of ORs. We also discuss the challenges and propose directions for future research in odor prediction and biosensor development. Sufficient and accurate experimental data are needed to improve odor prediction. The effects of concentration on odor quality need to be elucidated. Prediction models for odorant mixtures need to be developed by considering interactions among odorants. Binding affinity of odorant to OR is a potential input feature for odor prediction. • Data sources for molecular, receptor, and olfactory information are summarized. • Network of odorant – olfactory receptor – odor is mapped with machine learning. • Odorant concentration is a potential feature for odor quality prediction. • Application of binding free energy of odorant and olfactory receptor is proposed.