top of page
  • Writer's pictureTriple Helix

Got Your (E)-Nose!: Ai & Machines Are Learning to Detect Smell

Written By: Sara Santacruz ‘25

Edited By: Samuel Lew ‘24

If you were to ask anyone which of their five senses they would be the most okay with losing, most would choose to give up their sense of smell. Suddenly losing the ability to see, hear, or touch the surrounding world sounds like a challenge. The loss of smell, however, might even come as a blessing- no need to hold your nose as you make a quick stop at the communal bathrooms!

Researcher Guangyu Robert Yang and his team at MIT more recently brought the importance of smell back into the scientific spotlight by creating an artificial olfactory system. In collaboration with Columbia University neuroscientists Richard Axel and Larry Abbot, the research team used a computer program to observe whether or not Artificial Neural Networks (ANNs) mimicking the olfactory system would organize themselves in a way similar to the biological systems of animals by providing it with a certain species’ neuron count and instructions to classify odors [1]. Once the program had been instructed to detect and classify the presented odors, it was discovered that the artificial olfactory system created to perform this task was nearly identical to the biological olfactory system of fruit fly, even down to its number of compression layer connections: 6 total [1].

The results of this experiment open up a new method of studying the complex and fragile brain by utilizing artificial intelligence. Yang’s team, for example, produced a computer program trained to classify odors by both their class and valence and found that unique characteristics such as a three-layer input convergence expansion structure are necessary to complete such a specific task [1]. Beyond being able to identify these trained odors, the program used this precisely chosen artificial olfactory system to learn how to distinguish novel odors from each other. Such a study seems to provide evidence that the evolution of the brain’s biological structures, such as the olfactory system, are intentionally organized with the goal of providing the most optimal results in odor classification.

Not only does such an experiment provide insight into the purposeful design of various brain structures, but also creates an interesting dialogue about how smell in Artificial Intelligence can be used to fill specific niches in fields such as medicine. In an Electronic Nose, the ‘olfactory system’ is composed of a gas sensor array that ‘reads’ odor by converting its gas molecular signals into electrical signals [2]. Several scientists find that the advanced sense of smell held by Electronic Noses has the capability to break down and fully analyze more chemically complex human characteristics such as breath [1]. An individual’s breath is as unique to them as their fingerprints, containing chemicals that have the potential to distinguish between a healthy and an ill person during the early stages of a disease.

It is not unusual to notice you have a distinct smell when you are sick. Throughout history, multiple diseases have been attributed to specific odors that help in identification, including the 1970s yellow fever, which was described as causing a butcher-shop-like smell to follow the infected [3]. In a similar manner, Electronic noses can identify a disease by analyzing the breath of a patient and its chemical structure, locating its distinct biomarkers. In fact, the analysis of these biomarkers is not just useful for surface-level disease identification, as Electronic noses are also able to use them to detect diseased tissues at specific locations within the human body [3].

It costs no extra energy on the part of the patient to breathe in and out. The reader of this article has probably exhaled unconsciously multiple times without even noticing it. For such little cost, it is almost shocking to know exactly how valuable the breath is in diagnostics. The exhaled breath of a healthy human being contains numerous amounts of volatile organic compounds (VOCs) that are normally undetectable when the body is in perfect condition. In the case where an individual feels slightly off or has caught a disease somewhere within their body, abnormal VOCs that are even less detectable become present, yet are still readable by incredibly strong technology such as the Electronic Nose [3]. Such facts are part of why Electronic noses are becoming attractive not only as a point of the future of neurological study, but as a diagnostic tool in the medical industry. This novel technology is inexpensive, easy to use, and most importantly non-invasive to the patient, posing the olfactory system as the unlikely secret weapon of medicine.



[1] Wang PY, Sun Y, Axel R, Abbott LF, Yang GR. Evolving the olfactory system with machine learning. Neuron [Internet]. 2021;109(23):3879-3892.e5. Available from:

[2] Ye Z, Liu Y, Li Q. Recent progress in smart electronic nose technologies enabled with machine learning methods. Sensors (Basel) [Internet]. 2021 [cited 2022 Dec 1];21(22):7620. Available from:

[3] Wilson AD. Future applications of electronic-nose technologies in healthcare and biomedicine. In: Akyar, Isin, ed 2011 Wide Spectra of Quality Control InTech Publishing, Rijeka, Croatia 267-290 [Internet]. 2011 [cited 2022 Dec 1];267–90. Available from:

(IMAGE SOURCE) Universitat Rovira i Virgili. Science in 1 minute: How does an electronic nose work [Internet]. Youtube; 2016 [cited 2022 Dec 10]. Available from:

7 views0 comments


bottom of page