The technology teaching machines to smell

The simplest way to determine whether milk has gone off is to give it a sniff. Unfortunately, this isn’t yet something computers have learned to do, which means smart fridges aren’t clever enough to tell us that we need to visit the dairy aisle at the supermarket.

However, this may one day be possible. Professor Krishna Persaud of the University of Manchester is one of the scientists currently looking for ways to teach machines to smell – and he’s just published a breakthrough piece of research about smart sensors.

An interview with Professor Krishna Persaud

Your research has found a way to recreate proteins normally found in the nose. So have you effectively replicated our sense of smell?

They’re not actually sensors or receptors; they’re proteins that are carriers for smells. We’ve found a way to mass-produce these proteins using molecular genetics techniques.

We’ve also discovered how to modify these proteins in such a way that they can bind with compounds with which they wouldn’t normally. So, we can actually target them. In this instance, we’ve chosen to modify a protein to detect molecules that are mirror images of themselves. We’ve used a compound called carvone, which has two forms: one smells like spearmint and the other smells like caraway. They’re the same molecule chemically – they’re images of each other.

We can actually make this protein detect these two compounds easily. The next step was to try to get a signal from these proteins, if we were to make a biosensor.

How was this achieved?

The proteins on their own don’t do anything, they’re only responsible for binding a molecule. We – in collaboration with an expert group at the University of Bari, led by Professor Luisa Torsi – have placed these proteins onto the gate of the transistor, so that when a molecule binds to that protein, we can receive a signal from the transistor.

And that potentially gives computers the ability to smell?

This signal is actually a change in the current that passes through the transistor. So that when a molecule binds, we can receive a simple signal to measure it. We can detect, say for example, 50 picomolar – that’s 50 molecules in a billion molecules.

How could these biosensors be used in the real world?

One of the areas we had in mind was food-quality assurance. For example, smart labels in packaging, where we have the ability to actually follow sensitive goods all the way from the manufacturing process, through transit, to the customer, to ensure that they’re in good quality when they reach the customer.

The other significant application is environmental monitoring. The advantage that we have using this kind of device is that biosensors actually have very, very short lives: they’re usually used once and then thrown away. In this case, we can actually use a sensor for a considerable amount of time, which means that we can follow events in real-time. So if you think of, say, monitoring pollution in the environment, then this allows you to deploy sensors and monitor target pollutants in real-time.

Could smart sensors and labels be put to use in a smart fridge, to let you know that your milk has gone off?

Yes, these sensors cut to the heart of such devices. I think it’s going to be a few years yet, because… [this is] really a proof of concept to show that this approach actually works.

Interested in finding out more? Click here to read the research paper by Professor Persaud and his colleagues. One word of warning: make sure you have your science hat on; this is serious stuff.