In the realm of neuroscience, a fundamental aspect involves unraveling the mechanisms by which our sensory perceptions translate various stimuli into conscious experiences – be it light into sight, music into hearing, food into taste, or texture into touch. However, the intricate relationships governing our sense of smell have long puzzled researchers.
The human perception of odors, whether delightful floral scents or offensive odors from decaying food, is influenced by proteins in the nose known as odor receptors. Yet, our understanding of how these receptors detect and convert chemicals into fragrances remains limited.
A team of researchers from the Monell Chemical Senses Center collaborated with the Cambridge-based startup Osmo, an initiative stemming from Google Research’s machine learning efforts to shed light on this phenomenon. The scientists embarked on a journey to explore the intricate connections between the brain’s olfactory perception system and airborne chemicals. Their study culminated in the development of a machine learning model capable of verbally describing the scent of compounds with a level of proficiency comparable to humans.
The details of their groundbreaking study have been published in the journal Science.
Extensive efforts
This research endeavor required extensive effort and attention. Humans possess approximately 400 active olfactory receptors. Significantly, this number exceeds the four receptors employed for color vision or the 40 used for taste perception. All these receptors interact with airborne chemicals to send electrical signals to the olfactory bulb.
Joel Mainland, a senior co-author and a member of the Monell Center, stated that “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.” The research team diligently worked to elucidate the relationship between molecular structure and the perception of odors.
To achieve this, the team constructed a model capable of correlating written descriptions of a molecule’s odor with its molecular structure. The resulting map clustered scents with similar aromas, such as “floral sweet” and “candy sweet.”
The machine was trained using a comprehensive dataset containing the molecular makeup and olfactory characteristics of 5,000 recognized odorants. The algorithm took the molecular structure of a compound as input and predicted the most suitable odor descriptors.
To validate the model’s efficacy, researchers at Monell conducted a blind validation process involving a panel of trained participants. These individuals were tasked with describing new molecules using a set of 55 words ranging from “mint” to “musty.” Their descriptions were then compared with those generated by the machine.
Outstanding results
The results were impressive. The AI model outperformed each panelist in describing 53% of the compounds studied. Moreover, the model excelled in olfactory tasks it had not been specifically trained for. Joel Mainland noted that “the eye-opener was that we never trained it to learn odor strength, but it could nonetheless make accurate predictions.”
The program demonstrated the ability to quantify various odor attributes, including odor intensity, for 500,000 potential scent molecules. It also uncovered numerous instances where structurally dissimilar compounds shared surprisingly similar odors. Mainland also expressed hope that this map would serve as a valuable tool for researchers in chemistry, olfactory neuroscience, and psychophysics, facilitating the exploration of the nature of olfactory sensation.
The team speculated that this model could potentially be organized based on metabolic pathways. This could further mark a significant shift in how scientists categorize and understand scents. Currently, sensory scientists classify compounds in the manner of chemists, considering factors such as the presence of esters or aromatic rings.
The implications of this study are vast. It may bring the world closer to digitizing and reproducing odors, potentially revolutionizing industries like fragrance and flavor. Additionally, it could identify new odors for various applications, from reducing dependence on endangered plants to developing functional scents for purposes like mosquito repellent or malodor masking.
What lies ahead for the team? Their next objective is to delve into how odorants interact and compete with each other to produce distinct aromas that the human brain perceives as entirely different from each individual odorant.
