This article was published on October 24, 2019

Google is training an AI to predict a molecule’s smell


Google is training an AI to predict a molecule’s smell

With plenty of mics and cameras at disposal, AI has gotten good at ‘seeing’ and ‘listening.’ But one human sense it hasn’t got around much is smell. Now, researchers at Google are trying to develop a neural network that helps an AI identify the smell characteristics of a molecule.

The company said identifying smell is a multi-label classification problem, meaning a substance can have multiple smell characteristics. For instance, Vanillin, a substance often used to create an artificial vanilla flavor, has multiple smell descriptors such as sweet, vanilla, and chocolate, with some characteristics stronger than others.

So, to identify the smell profile of a molecule researchers used a graph neural networks (GNNs), a deep learning model that takes graphs as inputs. The team took the help of perfume experts to create labels of smell that can be used to identify a molecule’s olfactory properties. 

The neural network starts the process by creating a representative vector using various properties such as atom identity and atom charge. Then it broadcasts the vector to a neighboring node, and then collectively passes to update function to get a vector for centered node. 

Each node is represented as a vector, and each entry in the vector initially encodes some atomic-level information.

This process is repeated for a layer, and then it continues for multiple layers. Finally, the AI sums up or averages a vector for a molecule to identify multiple olfactory identifiers. 

Illustration of a GNN for odor prediction. We transform molecules into graphs with information and these are fed into GNN layers to learn a better representation of the nodes. These nodes are reduced into a single vector and passed into a neural network that is used to predict multiple odor descriptors.

Google researchers said not only this model outperforms older methods, but it can be used to predict new or unclassified smells in RGB-layout like “odor embedding”.

In the future, the team wants to create solutions for digitalized scent creations and even build solutions for those without a sense of smell. Plus, it wants to create more open datasets for research so researchers can leverage them for various scent-related machine learning models.

Get the TNW newsletter

Get the most important tech news in your inbox each week.

Also tagged with