PNAS: A surprisingly 'simple' algorithm for smell

Smell it indoors or out; summer or winter; scones at the coffee shop; pepperoni at the pizzeria — even scones at the pizzeria! Coffee smells like coffee
.

Why don't other odors or different environmental factors "get in the way," that is, the experience of smelling individual odors? Researchers at Washington University's McKelvey School of Engineering in St.
Louis turned to their trusty study subjects, the locust, for answers
.

Barani Raman, a professor of biomedical engineering, said their findings were "really simple"
.
Their findings were published in the Proceedings of the National Academy of Sciences

Raman and his colleagues have been studying locusts for years, looking at their brains and their smell-related behaviors, trying to engineer bomb-sniffing locusts
.
In the process, they have made significant progress in understanding the olfactory mechanisms of locusts

To understand how locusts consistently recognize scents in any environment, they took a cue from Ivan Pavlov
.
Like Pavlov's dogs, locusts are trained to associate scent with food, and their preference is a piece of grass

At this point, the researchers began to observe when locusts were exposed to the odor under different conditions (including with other odors), in wet or dry conditions, when hungry or full, in trained or untrained conditions time, and which neurons were activated at different times
.

The results showed that, in different settings, the researchers found that even when the tentacles of the locusts turned on each time, the neuronal patterns of activation were highly inconsistent
.
"Neural responses are highly variable," Raman said

How do variable neural responses produce consistent or stable behavior? To explore this, the researchers turned to a machine-learning algorithm

This algorithm is easy to explain

"You can think of ON neurons as transmitting 'evidence' for the presence of an odor, and OFF neurons as transmitting 'evidence' against the presence of an odor," Raman said
.
To identify the presence of an odor, the researchers only need to add the odor.

"We were surprised to find that this simple method was all that was needed to identify an odor," Raman said
.

Raman likens the process to buying a shirt
.
Let's say you have a list of qualities you want - cotton, long sleeves, buttons, solid colors, maybe a front pocket for glasses - plus a few must-haves , such as dry-clean only or polka-dotted clothes

You may be lucky enough to find a shirt you are looking for
.
But, more practically, you'll buy as soon as many of the features you want appear and most of the ones that don't qualify don't
.

Look for the features you want that are similar to the information conveyed by ON neurons
.
There is no break-up mechanism similar to turning off neurons
.
As long as enough ON neurons that are normally activated by smells fire—and most OFF neurons don’t—then it can be safely predicted that locusts will open their tentacles in anticipation of eating grass
.

Journal Reference :

1. Srinath Nizampatnam, Lijun Zhang, Rishabh Chandak, James Li, Baranidharan Raman.
   Invariant odor recognition with ON–OFF neural ensembles .
   Proceedings of the National Academy of Sciences , 2022; 119 (2): e2023340118 DOI: 10.
   1073/pnas.
   2023340118 nuary 10 , 2022).