How do our senses work? The way our brains interpret and understand smells, sounds, tastes, and other sensory inputs has long troubled researchers. However, a new model could bring scientists closer to the digital replication of our senses
A core focus of neuroscience lies in understanding how our senses turn light into sight, sound into hearing, food into taste and texture into touch. Smell introduces heightened complexity and intriguing intricacies to these sensory interactions
A research team co-led by the Monell Chemical Senses Center and start-up Osmo, a Cambridge, Mass-based company, originated from machine learning research at Google Research (formerly known as Google Brain) and Google DeepMind, is studying how airborne chemicals connect to odour perception in the brain and how technology can be used to describe sensory inputs in greater detail than ever before.
Understanding odour perception in the brain
By creating a machine learning model the team have discovered the ability to describe how chemicals might smell in words. Their research appears in the September 1 issue of Science.
Essentially the model helps explain how our brains and noses work together everyday to understand the world around us in different ways.
“The model addresses age-old gaps in the scientific understanding of the sense of smell,” said senior co-author Joel Mainland, PhD, Monell Center Member.
Mapping Odours
This collaboration moves the world closer towards the digital replication of our senses, especially smell.
In humans, there are approximately 400 operational olfactory receptors. These proteins at the end of olfactory nerves connect with airborne molecules to transmit an electrical signal to the olfactory bulb. This number of olfactory receptors is significantly greater than we have for colour vision (four receptors) or taste (around 40).
“In olfaction research, however, the question of what physical properties make an airborne molecule smell the way it does to the brain has remained an enigma,” said Mainland.
To tackle this challenge, Osmo CEO Alex Wiltschko, PhD, and his team developed a model that links written descriptions of how a molecule smells with its actual molecular structure. This led to a map grouping similar-smelling odours, such as floral sweet and candy sweet.
“This study proposes and validates a novel data-driven map of human olfaction, matching chemical structure to odour perception,” said Wiltschko.
Metabolic insights into the digital replication of our senses
The model was tested using an industry dataset containing information about 5,000 known odour-causing substances, including their molecular structures and odour characteristics.
The input for the model is the shape of a molecule, and it predicts which words best describe how it smells.
To test the model, researchers had 15 trained participants describe new molecules and then compare these with the model’s generated descriptions. Each panellist had 400 odorants to describe, using a set of 55 words that ranged from “mint” to “musty” to characterise each molecule.
Co-author Jane Parker, PhD, a Professor of Flavor Chemistry at the University of Reading, UK, played a key role in the quality control of the research. Parker’s team ensured the purity of the samples used to test the model’s predictions.
They accomplished this by using gas chromatography to separate the compounds in each sample, including any impurities. Subsequently, Parker and her team smelled each separated compound to check if any impurities were overpowering the intended odour of the target molecule.
When comparing the model’s performance to individual panellists, the model achieved better predictions of the average group odour rating, excluding impurities.
The model achieved better predictions of the average group odour rating
The model successfully identified many pairs of molecules with different structures but surprisingly similar smells. It also described various odour properties, including odour strength, for a whopping 500,000 potential scent molecules.
