An electronic system has been trained by the scientists from the Weizmann Institute for predicting the pleasantness of new odors, similar to humans who perceive them.
The PLoS Computational Biology journal has published this research. The scientists informed that the human perception that is related to the odor’s pleasantness is hard-wired innately to its molecular structure and the cultural or personal differences among humans in perceiving the pleasantness of an odor are apparent only in specific contexts. This runs contrary to the widespread notion that smell is totally cultural and person specific.
The research findings are important for developing applications for monitoring the toxicity in the environment, quick screening of the odor for the perfume industry, and malodor monitoring. These findings will help to create a vital building block for a future road map of the sense technology domain ultimately to transmit the odor digitally.
Electronic gadgets, also termed as 'eNoses,’ or ‘electronic noses’ were developed during the past decade for detecting and recognizing odors. The chemical sensors array is the key element of an eNose. As an odor passes through the electronic nose, its molecular features are made to stimulate the sensors in a manner as to generate an odor finger print that has a special electrical pattern and characterizes a particular type of odor. This nose has to be trained using samples, like a sniffer dog, for generating a reference database through which this instrument will be able to recognize more samples due to such odors by comparison of the fingerprint. However, if the new odor is not present in this database the noses will not be able to identify and categorize it.
This issue prompted the Weizmann scientists to tackle this issue from a different perspective. They trained the nose to predict the odor on a specified axis of perception, rather than training it to identify a certain odor. Odorant pleasantness was the axis chosen by them. They were able to train the noses to estimate if any odor could be perceivable as unpleasant or pleasant or mid-way between the two.
They used Israeli respondents for rating the degree of pleasantness of a range of odors and created a dataset that was made using a 30-point scale that varied from very unpleasant to very pleasant. Using this dataset they created an algorithm of odor pleasantness that was subsequently programmed into the electronic nose. The nose was then made to estimate the novel odors for their respective unpleasantness feature that was not recorded in the database and compared them with the ratings provided by another set of respondents. The nose could now generalize the odors as well as rate them, and the ratings were found to be similar to the extent of 80% to those of respondents who were not part of the earlier training phase. It was possible to achieve 99% accuracy if the odors were merely classified as pleasant or unpleasant, rather than rate them based on a scale.