A human being senses a smell when molecules ring the sensory receptors present in the nose. It is harder to predict smell than color. Because, unlike the human eye, which has only three sensory receptors for sensing the photons of red, green and blue color, a human nose has over 300 receptors.  

Therefore, there is no argument that sensing smells is a difficult task. But what if an AI model could detect various odors? 

Google has built an AI model with a human-like capacity to predict odor. The map developed by the team of Google AI links molecular structure to the aroma of substance and can even predict smell that is still unnoticeable by humans. 

The Research 

In 2019, scientists commenced exploration with a deep learning algorithm. It was an interplay of molecular structure with the type of smell. A graph neural network (GNN) model was trained to identify various examples of particular molecules paired with the smell labels they arouse, such as beefy, floral or minty. 

With this research, scientists have successfully formulated a “Principal Odor Map” (POM) with the properties of a sensory map. Furthermore, they tested the model on various parameters.  

The scientists tried to find out if the model has learned to predict the odors of new molecules that humans have never smelled and differed from molecules to train the GNN model. In the Google post, the researchers called the study an ‘important test’. Many models perform well on data that looks similar to what it has seen before but break down when tested on novel cases. 

Google’s model stood on the test and presented exceptional intelligence to predict smell from the structure of the molecule.  

The scientists also tested if the model could predict odor perceptions in animals. According to their attempt, they found that the map could well predict the activity of sensory receptors, neurons and behavior in most animals that olfactory neuroscientists have studied, including mice and insects. 

Google’s team found that the common purpose of the ability to smell might be to detect and discriminate between metabolic states, i.e., to sense when something is ripe vs rotten, nutritious vs inert or healthy vs sick.  

They had gathered data about metabolic reactions in dozens of species across the kingdoms of life and found that the map corresponds closely to metabolism itself.  

Application of the research 

The scientists retrained the model to tackle the issue of the spread of diseases transmitted by mosquitoes and ticks while killing hundreds of thousands of people each year.  

The team improved the original model with two new sources of data. The first set was a long-forgotten set of experiments conducted by the USDA on human volunteers beginning 80 years ago and recently made discoverable by Google books. Secondly, a new dataset was collected by their partners at TOPIQ, using their high-throughput laboratory mosquito assay.   

Both datasets measure how well a given molecule keeps the mosquitos away. Together, the resulting model can predict the mosquito repellence of nearly any molecule, enabling a virtual screen over huge swaths of molecular space. 

Moving Afore 

From their study, the researchers found out that their approach to smell prediction could be used to draw a Principal Odor Map for tackling Odor-related problems more generally. The map was the key to measuring smell. It answered a range of questions about novel smells and the molecules that produce them. In addition, the model connected the smells back to their evolution and the natural world.