For the first time in the country, an algorithm has been developed by Central Scientific Instruments Organisation (CSIO) Chandigarh based scientists that allow predicting smell of new molecules. This basic science can have vast applications ranging from enabling smell of recipes to replicating natural odours.
For the first time in the country, an algorithm has been developed by Central Scientific Instruments Organisation (CSIO) Chandigarh based scientists that allow predicting smell of new molecules. This basic science can have vast applications ranging from enabling smell of recipes while watching cookery show or get the taste of exotic food while retaining the smell of a traditional cuisine.
For the scientists, smell has been enigmatic as it has been the least understood of all our senses. So far chemical and physical properties could be predicted from molecular structure, however smell being a sense which is learnt by association, could not be predicted.
“It is all perception driven unlike other senses. Among other sensory perceptions including sight and touch there is a well-defined scale, but not so in smell. For instance, in hearing there is mapping of sound in terms of decibel. Likewise, light can be defined by hue, brightness, and saturation. In other words, it has been a challenge to predict the smell of a novel molecule by its physicochemical structure,” Amol P Bhondekar, principal scientist at the CSIO who worked on this project told Indian Science Journal.
What is the significance of this data mining software? Scientists claim the exploitation of natural resources for extraction of essential oils is evident from the fact that musk deers are on the verge of extinction now. “This calls for the need to create synthetic materials for replicating these natural odours thereby reducing the load on endangered flora and fauna,” said Amol.
The work on the algorithm started in 2012 funded by Council of Scientific and Industrial Research. The team included Ritesh Kumar, Rishemjit Kaur under the guidance of Bhondekar and Dr G P S Raghava. How was this data based formed? This team mined open access data of 3016 molecules, which included physcio-chemical properties of 1666 molecules and 526 perceptual descriptors (how preceptors that smelled the molecules defined it). Using this data, a model has been prepared that can predict smell of new molecules which are formed after they are blended.
However the main work started in 2015 with the Rockefeller University asking 49 volunteers to assess the odour of 476 different chemicals, based on 21 different descriptors. The descriptors ranged from a chemical’s intensity and pleasantness of smell, to how fruity, spicy or fishy it was. They later released it to open community of researchers to challenge them to design machine learning algorithms. CSIO was the only participant from India and was ranked 15th out of 22 finalists.
