Patterns and proofs

Patterns and proofs

Predicting what your food smells like just got easier

Not to be confused with a glass of wine. Mortefot

Familiar everyday odours such as coffee and red wine are produced by a blend of different substances. Given that we know aroma is nothing but a mix of volatile chemicals, can we understand them enough to predict the smell they will produce?

Our sense of smell is driven by an “olfactory system”, but scientists have remained puzzled as to exactly how it works. Being able to crack this nose code would help reveal how the brain processes scents. It may also bring riches to companies that can exploit such a technology.

That technology may not be far off, if researchers at the Weizmann Institute of Science in Israel are correct. Last month, they published a study in the journal PLOS Computational Biology investigating how molecules create a perception of smell.

They started with by treating an odour as the sum of its parts, which is how most electronic noses work. Looking up each constituent chemical in a “molecular encyclopaedia”, the Weizmann Institute team collected hundreds of different physical and chemical properties for each substance in their test odours. They then developed a range of aroma mixtures: some consisted of just four different chemicals; the most complex was made up of 43 different types of molecules.

With the odours ready, the researchers recruited a group of 139 volunteers, all of whom were “normosmic” - that is, they had a normal sense of smell - and got them to rate how similar or different pairs of smells were. Then came the tricky part: working out how to take the molecular structure of each pair of scents and translate it into a measurement of perceived similarity.

To measure the difference between two particular smells, they summed up the differences between the individual properties of the types of molecule they contained. So to compare a smell made up of two kinds of molecule with each made up of three molecule, they had to look at six differences (shown by arrows):

Adapted from Snitz et al. (2013), PLOS Computational Biology

When they compared these measurements with the comparisons reported by the volunteers, however, the results didn’t line up. It appeared the researchers couldn’t predict a smell by looking at differences between individual molecules. But perhaps it wasn’t reasonable to assume the participants’ noses analysed scents as separate molecules. After all, life doesn’t come as a series of individual smells; it usually involves a burst of them.

The researchers therefore decided to treat each aroma as a single “odour object”. To compare two aromas, they calculated a single measurement: the overall difference between the total physical and chemical properties of the first aroma mixture and the collective properties of the second.

Adapted from Snitz et al. (2013), PLOS Computational Biology

When they measured differences in this way, they had more success predicting how people perceived the smells. The results agree with other studies, which have shown people find it difficult to tease apart the different components of an odour. Researchers have found that given an aroma made from a mixture of substances, participants struggle to pick out more than four separate components.

What the Weizmann Institute researchers could not do is compare every possible smell in the study. Although they looked at several different combinations of odours, the range of potential smells in nature – known as the “olfactory perceptual space” – has not been explored in depth.

One study, published in PLOS One recently, suggested that the space might be 10-dimensional. Each dimension represented a different odour quality - such as fruity, minty or pungent - and real-life smells could be described by a specific combination of ten basic smells.

The researchers from the Weizmann Institute also carefully controlled the intensity of each odour they added to a mixture. In reality, however, certain scents might dominate the overall aroma, or even grow in intensity when mixed with other substances. Such complexity could affect the predictive power of the researchers’ method.

Even so, the technique was capable of analysing fairly complex aromas – some made up of dozens of different molecules – and converting the analysis into a “measurement” of smell. In doing so, the researchers have helped address some of the question of how our brains interpret complex odours such as coffee and wine – and why we sometimes think one smells like another.