The human sense of taste is the result of millennia of evolution. And its astoundingly good at letting us enjoy pleasant foods and beverages as well as warning us against ingesting harmful substances. Man-made sensors, on the other hand, have yet to approach the ease with which our taste buds recognize substances. This is a significant technological gap, as there are many substances out there that we would like to “taste” without actually putting them in our mouth. For the rapid and mobile fingerprinting of beverages and other liquids less fit for ingestion, our team at IBM Research is currently developing Hypertaste, an electronic, AI-assisted tongue that draws inspiration from the way humans taste things.
Hypertaste fills a gap in chemical sensing
The Hypertaste app.
Hypertaste caters to a wide range of industrial and scientific users with a growing need to identify liquids swiftly and reliably without access to high-end laboratories. Consider a government agency interested in an on-the-fly water quality check of a lake or river at a remote location. Or a manufacturer wanting to verify the origin of raw materials. Or a food producer trying to identify counterfeit wines or whiskeys. The quick, in-situ identification and classification of liquids is relevant also in the pharmaceutical and healthcare industries, to name just a few more examples.
Capabilities offered by present-day instruments for chemical sensing are either very specialized, moderately priced portable sensors to measure specific properties like pH, or, on the other end of the spectrum, very powerful stationary machines for the precise analysis of individual molecular components. Closing this gap is crucial as most liquids of practical use are complex, meaning they comprise a rather large number of chemical compounds, none of which can serve as an identifier alone. In these liquids, it’s not so much the single components that matter but rather the properties that arise from combining them. And yet, routinely sending such liquids to a lab for analysis is costly, time-consuming and often impractical. This is where Hypertaste comes in.
Minimalistic hardware thanks to combinatorial sensing
Since complex liquids contain so many different molecules, it would be inefficient to identify them by sensing each component separately. Hypertaste therefore uses combinatorial sensing instead. In that respect, it resembles our natural senses of taste and smell where we don’t have a receptor for each molecule occurring in every kind of food or drink. Combinatorial sensing relies on the ability of individual sensors to respond simultaneously to different chemicals. By building an array of such cross-sensitive sensors one can obtain a holistic signal, or fingerprint, of the liquid in question.
Hypertaste uses electrochemical sensors comprised of pairs of electrodes, each responding to the presence of a combination of molecules by means of a voltage signal, which is easy to measure. The combined voltage signals of all pairs of electrodes represents the liquid’s fingerprint. Key to the functioning of our electrochemical sensors are polymer coatings covering each electrode. At our lab in Zurich, we synthesize these coatings which are designed to capture a range of chemical information and allow a high degree of miniaturization.
Less than a minute from measurement to identification
We have built sensor arrays and combined them with off-the-shelf electronics that we configure to measure the voltages across the electrodes in an array and relay them to a mobile device, such as a smartphone. A mobile app transfers the data to a cloud server, where a trained machine learning algorithm compares the digital fingerprint just recorded to a database of known liquids. The algorithm figures out which liquids in the database are most chemically similar to the liquid under investigation, and reports the result back to the mobile app.
This type of task is called classification, and in our proof-of-concept the whole process takes less than a minute from the moment the sensor is dipped into a liquid to the display of the classification result on the mobile device. An important aspect of achieving this feat is the training of the sensor. Like a budding sommelier learning the intricacies of wine tasting, the Hypertaste sensor array needs to be trained to identify the liquids of interest before being put to the test. This is done by measuring the sensor array response in those liquids multiple times and then feeding the resulting data into a machine learning model which extracts the characteristic features associated with each liquid.
The portability and speed of the Hypertaste sensors is achieved by using combinatorial sensing with an array of cross-sensitive sensors combined with intelligent software that can be outsourced to the cloud. One big advantage of having the machine learning models running on the cloud is that the sensors can be rapidly reconfigured from anywhere without alterations to the hardware. All that is needed to “rewire” the sensors is change the parameters of the machine learning models to make them adjust to a new task. Sensors could learn from one another by exchanging information about new liquids they encounter. Deploying many such sensors in the field would add an important but missing building block to the Internet of Things: chemical sensors.
How Hypertaste adds value
Hypertaste proves that a portable device could be capable of rapid fingerprinting of complex liquids – a capability currently lacking in the toolkit of chemical analytics. Industries and services that would benefit from such a technology range from industrial supply chains, food and beverages and environmental monitoring to the pharmaceutical and healthcare sectors, to name just a few.
As an example of the advantages of the Hypertaste combinatorial sensing approach, think of the supply chain safety from producer to consumer for packaged food and drinks. At present, once food and drinks are packaged, there is little ability to verify that the package actually contains what is on the label, apart from sending the product to a lab for testing. So, suppliers acting in bad faith may insert lower-quality products into the supply chain with little risk of getting caught, or counterfeiters may even fake a real product by adding the few analytes which are most likely to be tested for in a lab. Fooling a combinatorial sensing system such as Hypertaste is much harder as there is no single substance on which the identification relies, and it is more difficult for wrong-doers to access the sensor training parameters which provide the “key” to interpreting the chemical fingerprints.
In the long term, we also envision using Hypertaste in fingerprinting even more challenging liquids, like those occurring in the life sciences. Sampling a person’s urine, for example, could help provide an assessment of the metabolomic fingerprint, which may be understood as the sum of all small molecules present in a living organism. As this rich chemical information is constantly changing depending on factors such as lifestyle and nutrition, metabolomic fingerprinting can be thought of as creating a snapshot of a person’s health.
Besides possible applications in diagnostic or preventive medicine, such a tool could also allow sub-grouping of patients in clinical trials for new drugs by matching the individual responses of patients to a treatment with information on their personal metabolomes. The spectrum of possible applications is vast and spurs the imagination. We are confident that, through forthcoming refinements, the use of AI-assisted, portable chemical sensors will meet the needs of many industries when it comes to the fast and mobile fingerprinting of complex liquids.
Founded in March 2020 just as the pandemic’s wave was starting to wash over the world, the Consortium has brought together 43 members with supercomputing resources. Private and public enterprises, academia, government and technology companies, many of whom are typically rivals. “It is simply unprecedented,” said Dario Gil, Senior Vice President and Director of IBM Research, one of the founding organizations. “The outcomes we’ve achieved, the lessons we’ve learned, and the next steps we have to pursue are all the result of the collective efforts of these Consortium’s community.”
The next step? Creating the National Strategic Computing Reserve to help the world be better prepared for future global emergencies.