AI picked up 40,000 chemical weapons in six hours

(ORDO NEWS) — Scientists at Collaborations Pharmaceuticals have developed a machine learning model for drug discovery. In the new work, they changed the methodology to identify potentially dangerous molecules.

The artificial intelligence algorithm was able to pick up 40 thousand variants of deadly molecules. They can become the basis for chemical weapons. Study lead author Fabio Urbina of Collaborations Pharmaceuticals shared the results of the work published in Nature Machine Intelligence.

Artificial intelligence has been developed to search for new drugs. It uses datasets of tested molecules and determines the efficacy and safety of substances. The prepared dataset allows the machine learning model to determine the parts of the molecular structure responsible for toxicity. Generative models connect molecules in the direction that scientists set.

In the new work, the researchers decided to change the methodology so that the algorithm does not screen out, but finds toxic variants of substances. Within six hours, artificial intelligence has identified dozens of molecules that are dangerous to humans.

Among them were substances resembling VX, an acetylcholinesterase inhibitor developed in the UK in the early 1950s. Outwardly, the substance resembles motor oil, but has no taste and smell.

The enzyme acetylcholinesterase is used by neurons to regulate muscle function. The VX nerve agent disrupts the enzyme and paralyzes the diaphragm and lung muscles.

The researchers said that among the detected molecules were substances with greater toxicity than VX. The structures of the resulting substances resembled VX and other types of chemical weapons. At the same time, some were created on the basis of models of real poisonous substances that were not included in the prepared database.

Urbina emphasized that the machine learning model made a prediction of toxicity, and it is possible to accurately determine the danger of the resulting molecules only in laboratory studies. He assessed the likelihood of using generative models and datasets to create chemical weapons.

According to the researcher, this is possible in theory, but in practice, after obtaining a model of a dangerous molecule, it must be synthesized. Substances in chemical weapons are not easy to obtain: they are monitored and produced by a limited number of companies.


Contact us: [email protected]

Our Standards, Terms of Use: Standard Terms And Conditions.