The research aims to reveal the potential smell of a chemical combination by connecting a given scent back to the molecular structure. This is based on an initiative launched by Rockefeller University scientists and unusually they’ve taken to crowdsourcing in order to raise funds for the research. This is to come up with a mathematical model in order to forecast the scent a molecule will evoke.
The research is led by Leslie Vosshall. Professor Vosshall has spent several years examining odor perception in humans and insects. The research, the Professor states, will enable faster development of chemicals – everything from perfumes to insectides.
The research project began with assembling a team of volunteers. To gather some initial information the research group asked the subjects to sniff a specially curated set of molecules, based on chemicals contained in vials. The molecules were designed to open up the full spectrum of possible smells that a human can interpret. In all some 476 different scented molecules were selected. Many of these had not been subject to a smell study before.
The odors stretched from the sweet warmth of vanillin to the ghastly reek of methylthiobutyrate (a little like stinky-cheese). There were also molecules that most people would not have experienced before, such as 2-isopropylphenol. Each of the study participants rated each molecule based on:
How strong was it?
How pleasant is it?
What attributes does it evoke? (for which there was a list to select from, covering such terms as garlic, flower, urine and so on).
The entire exercise produced a massive one million data points. These were then entered into existing databases containing over 2 million data points covering the chemical features of the smell molecules.
This generated so much information the researchers took to crowdsourcing for the analysis. For this twenty-two teams of computer literate volunteers participated in the DREAM Olfaction Prediction Challenge, coordinated by IBM’s Thomas J. Watson Research Center. The teams worked to devise algorithms that could “learn” to predict an odor’s attributes based on a molecule’s chemical features.
This led to a workable set of models. None are absolutely perfect but they represent a significant advancement in the field. Those that were most accurate were models that could readily predict smells like garlic and fish.
The research has been published in the journal Science, under the heading “Predicting human olfactory perception from chemical features of odor molecules.”
