It is our oldest and most primitive sense, but also the least predictable one. While physicists and physiologists can easily explain the way our ears and eyes function, they are at a loss when it comes to our noses. Smells are so difficult to describe that researchers of olfaction face an immense challenge simply to find a standardize language to describe their object of interest. Quantifiable objective measurements of smells do not exist. There is a clear correlation between the wavelength of light and its color or the frequency of a sound and its pitch. No one knows, however, why a foul egg stinks or a rose smells so nice. So far, no link has been discovered between the molecular structure of a substance and its smell. The mechanism by which we perceive and interpret smell has therefore remained one of the largest mysteries of the human body. A team at Israel's Weizmann Institute now claims to have solved the mystery. Using a mathematical model, they have succeeded in predicting the smell of newly created substances, a first in the history of olfactory research.

Professor Noam Sobel and Dr. Rehan Khan of UCLA, authors of an article in the prestigious journal "Neuroscience", are sure that our perception of smells is genetically determined. Lab rats are terrified by the smell of cats even if they are inbred to the fourth generation and have never encountered their natural predator. Feline scents cause them to freeze in their place and release stress hormones into their bloodstream. In humans, the reaction to a smell seems to be instinctive as well, by and large. Sobel vehemently opposes the notion that humans have different tastes when it comes to smell. "In most cases," he says, "there is a baffling cross-cultural agreement on the question if a certain substance has a good or a bad smell." A Swede's predilection for eating rotten Hering or a Frenchman's enjoyment of ripe cheese are acquired tastes, claims Sobel. "We can learn that cheese is good for us and appreciate its smell. But even in France you will never find a perfume made to copy the scent of Roquefort or Raquelette."

Up to now, researchers have attempted to describe smells using a list of 146 attributes. To Sobel, a 40-year old neurobiologist with a penchant for statistics, this seemed too cumbersome. Using a statistical method called principal component analysis (PCA) he simplified the complex attributes and organized them on a continuous scale. The result surprised him. "When our brain encounters a new smell," he says, "it sorts it primarily according to how pleasant it is to us."

In the next step, Sobel applied PCA to the molecular structure of the fragrances. Chemicals are catalogued using a list of 1514 building blocks. Again, Sobel arranged molecules on a scale. Now his team asked volunteers to grade the smell of the substances in question according to their pleasantness, from stench to perfume. Then he correlated the results to his molecular scale.

Using this computed relationship as a staring point, Sobel advanced to the decisive step of his experiment. He asked his volunteers to grade the pleasantness of newly synthesized scents. Their answers correlated with his computed predictions. For the first time in history, a computer had predicted the pleasantness of a substance's smell.

It can be said that a small and simple molecule is more pleasant to the human nose than a large one that is densely packed. "But that accounts for only 40% of the smell," Sobel admits. To be able to make more accurate predictions, he now wants to create a data base of 1000 substances that will be sniffed by 150,000 volunteers.