New AI Model Can Identify Designer, Research Drugs on the Fly

A prestigious science award has been awarded to a man who has created a new generative AI language model that can identify the exact chemical structure of designer drugs, even those that have not yet been tested in humans.

The winning entry in the 2023 NOMIS & Science Young Explorer competition was a new AI language model trained by Princeton biologist Michael Skinnider. This new AI model is reportedly able to identify the chemical structure of research chemicals or “legal highs,” a term that refers to any number of compounds not yet planned by the FDA to produce psychoactive effects. The rapid rate at which these compounds are synthesized allows chemists and research institutions to sell dangerous drugs to people with labels that say “Not for human consumption,” more or less without fear of legal repercussions. A loophole in the law has been created that allows them to be shipped.

As a result of this loophole, law enforcement experts suspect that someone is in possession of a dangerous drug but cannot prove it, or that someone is suffering side effects from a drug but cannot be identified because they cannot be identified. Often faced with situations. Treat appropriately. While traditional field test kits cannot identify drugs because they only look for the most commonly used psychoactive compounds (heroin meth, cocaine, the usual suspects), Skinider’s AI model It has been reported that completely new chemical structures can be generated and identified in a hurry.

“While traditional drugs of abuse such as cocaine and methamphetamine still dominate the market, enterprising chemists are making small chemical modifications to these drugs to create new, completely legal derivatives. “We realized that it is possible to do this,” Skinnider wrote in his contribution. science. “And because these synthetic drugs have never been tested in humans, they can have unpredictable and harmful side effects.”

Skinnider began his work by training an AI language model with a simplified molecular input line input system, also known as SMILES. This is a fancy way of saying that you’ve taught your language model a new language that can be used to represent a variety of complex chemical structures. A simple text-based format.

Skinnider taught an AI model how to identify chemicals using a process called “mass spectrometry.” waters, Mass spectrometry is the process of measuring the ratio of different charges at the molecular level of something you want to test (in this case, a drug) in order to determine the exact molecular weight of the particles in the sample. These molecular weights are used to identify and map compounds.

“As an MD/PhD student, I saw firsthand how patients develop severe symptoms of designer drug addiction, but emergency physicians have few options for treating patients. “I thought artificial intelligence could be useful,” Skinnider said. “Specifically, we asked whether AI could automatically solve the chemical structure of new designer drugs from mass spectrometry data. Scientifically, this was a tall order.”

Skinnider then used information from existing research on commonly used designer drugs to further educate the AI ​​model using 1,753 known examples. What he discovered was that the program could generate entirely new examples of chemical structures that could have similar effects. Not only that, but it can also be used to predict which undiscovered chemicals are most likely to become popular in the future, based on how drug users have responded in the past.

Skinnider hinted that these advances in drug identification technology have very practical real-world applications in identifying and responding to drug crises. He also said that his technology is already being used to identify new and dangerous psychoactive compounds.

“I have now applied this technology to tens of thousands of patient samples and used it to discover several new designer drugs, including a new analog of fentanyl that emerged last year.” “We are working with the Centers for Disease Control to bring this AI technology into everyday clinical practice so that new drugs can be automatically discovered as soon as they are introduced to the population,” Skinnider said. “Ultimately, my dream is to enable first responders, emergency physicians, and public health officials to leverage generated AI to make more informed decisions in treating patients and managing outbreaks. It’s about becoming like that.