A team of Harvard scientists led by Venkatesh Murthy, professor of molecular and cellular biology, tackled a seemingly complex phenomenon in their investigation: how mice can distinguish scents from one another. Using a machine-learning algorithm, Murthy and his colleagues “trained” a computer to recognize the neural patterns linked to various scents, and to identify specific odors in a mixture of smells.
The study, published in the journal Neuron, focused on creating an algorithm for a computer to detect smells in a potentially similar fashion to mice. Picking out a neural activation pattern proved surprisingly simple, requiring only a simple linear classifier. The computer ended up working on a pattern-recognition system, in which patterns of neural activation (as a result of mice smelling certain odors) were registered.
As Murthy explained, “The computer looks at the patterns, randomly selects pixels, and adds them up,” Murthy explained. “If they reach a certain level, it says the target is there. Initially, though, it is almost certainly going to make a mistake. There’s then a process of fitting, in which we take the responses the computer gave and the actual responses, and we train it with the correct answers.”
After thousands of trials, Murthy said, the algorithm became as adept as mice at identifying the presence of a specific odor in a mixture of scents. Their findings suggest that mice may be employing a similar algorithm.
The scientists tested the computer to see if it could identify odors with exceedingly complex mixtures and discovered that, like mice, there a kind of detection limit that can be reached.
Not only did this research exemplify how mice are able to discern individual scents, the study also suggests that computer-learning algorithms can be powerful tools to examine olfaction, and could potentially be used to conduct experiments in a virtual space before conducting them in the real world.
