By akademiotoelektronik, 06/11/2022

DarkNPS, the artificial intelligence that has already imagined millions of new drugs

It's a bit like the "Minority Report" of the illicit substance.

Spotted by Thomas Messias on Undark

It's always a game of cat and mouse: the manufacturers of synthetic drugs are always one step ahead of the institutions and individuals whose role is in particular to determine the risks of substances on the psychological and physiological. When the existence of a new drug is discovered, its composition must be analyzed and the consequences of its absorption assessed, precious time is often wasted.

At the University of Alberta, considered one of the five most prestigious in all of Canada, the team of computer scientist and biologist David S. Wishart has been working on a way to reverse the trend and to gain the upper hand over the drug producers. The key, says Undark, is a deep learning artificial intelligence, capable of constantly assimilating new knowledge thanks to a network of artificial neurons. Its name: DarkNPS, the acronym NPS designating what is called in French the "new synthetic products".

Breaking Bad

Thanks to a database containing more than 1,700 existing drugs, DarkNPS has brainstormed to invent itself new products likely to be put on the market. Purely virtual creations, made so that they can be studied in advance, even before real manufacturers have the idea. In all, this AI has imagined no less than 8.9 million new synthetic products, with absolutely remarkable flair.

During the creative work done by Dark NPS, David S. Wishart and his colleagues (who just published the results of their work in the journal Nature) added 189 new drugs to their database. To discover with immense satisfaction that 176 of these recently invented products, or 93% of them, were among the predictions of their algorithm.

See alsoIn Silicon Valley, mini-doses of drugs to boost your career

Until now, the method used to identify the components of a new designer drug has been mass spectrometry, a method that can take weeks – or even months– before obtaining satisfactory results. It consists of determining the mass of each element forming the product in order to determine its molecular structure, then comparing the results obtained with the drugs appearing in a list of existing substances.

This technique is time-consuming and moderately effective, especially when compared to what DarkNPS is capable of accomplishing: “As soon as a drug hits the market, instead of three or six months of complicated chemistry, it is possible to identify its composition in a few minutes”, assures David S. Wishart.

A lot of work to do

DarkNPS' omniscience isn't total, however. For example, AI does not have the ability to provide information about the psychological or physiological effects of imagined drugs. Undark also evokes the danger that would represent the arrival of such a tool in the hands of malicious chemists, even if DarkNPS has obviously been made safe by the DEA (US federal agency responsible for combating drug trafficking and distribution) and the German Federal Criminal Police.

Read alsoOn the darknet, the drug market is going through a golden age

Another tool should therefore be developed to determine the effects of drugs created by DarkNPS on human beings. And keep in mind the fact that the algorithm is not an infallible genius: it relies on existing drugs to invent more or less similar substances, but is incapable of imagining totally different products. What some very talented chemists are capable of doing. Undark cites, for example, the case of a synthetic opioid that arrived on the market in 2020, isotonitazene, which DarkNPS clearly could not have invented.