Machine Learning Based Modelling of Human and Insect Olfaction Screens Millions of compounds to Identify Pleasant...

Area of Science:

• Computational neuroscience
• Chemosensation research
• Machine learning applications

Background:

• Predicting olfactory perception and valence of volatile chemicals is challenging due to complex neural processing.
• Previous work demonstrated chemical informatics for predicting odorant receptor ligands in Drosophila.
• Understanding the systems-level processing of olfactory information is crucial for rational odorant discovery.

Purpose of the Study:

• To develop machine learning models for predicting olfactory valence in insects and humans.
• To apply these models to explore a large chemical space for novel odorants.
• To identify compounds with differential valence across species for applications like insect repellents.

Main Methods:

• Utilized chemical structure-based machine learning models.
• Modeled olfaction in humans (GPCRs) and insects (ion channels) with distinct olfactory receptor proteins.
• In silico evaluation of over 10 million chemical compounds.

Main Results:

• Achieved high success rates in predicting human and insect behaviors.
• Successfully identified compounds with desirable human fragrance properties that are repellent to insects.
• Demonstrated the power of machine learning in predicting olfactory valence across species.

Conclusions:

• Machine learning provides a powerful tool for modeling complex olfactory systems.
• This approach facilitates the rational discovery of behaviorally active odorants.
• Offers a novel strategy for developing targeted insect repellents and fragrances.