Artificial intelligence to search for medicines came up with 40 thousand variants of chemical weapons
Artificial intelligence technology is no longer something new and unexpected. Today it is used in a wide range of areas and this is just the beginning. Scientists also use algorithms to find drugs.
As is often the case, technology has its own “evil” side. So, scientists from Collaborations Pharmaceuticals managed to invert the purpose of the megasyn molecule generator: instead of sifting out toxic substances during the search for new drugs, it began to specifically detect them. Fortunately, this is not enough to turn weapons into reality, because you also need special knowledge, equipment and raw materials.
Poison instead of Medicine
- It is reported that MegaSyn in just 6 hours of calculations offered 40 thousand options for potential biological weapons.
- The AI worked with a publicly available database of molecules that are mostly drug-like (i.e., those that can be absorbed by the body) and analyzed their biological activity.
- To narrow down the search, the researchers focused the algorithm’s attention on compounds such as the nerve agent VX, developed in the twentieth century.
- But MegaSyn went further. He developed compounds not only based on VX, but also on the basis of other well — known combat poisons – including Novichok, which poisoned former Russian spy Sergei Skripal and his daughter Yulia on March 4, 2018, and Russian opposition leader Alexey Navalny on August 20, 2020.
- Many new molecules have also been developed that are potentially even more dangerous than the already known ones.
I was concerned about how easy it was to do this. Much of what we used is available for free. You can download the toxicity dataset from anywhere. If you have someone who knows how to program in Python and has some machine learning skills, then probably […] it can build something like this generative model based on toxic datasets, – says Fabio Urbina, co-author of the study.
However, the researchers reassure: in order to use the model to the detriment, some specific knowledge is still needed. Without being an expert in the field, even with such a tool as AI, only inert material can be obtained at the output. In addition, molecules still need to be synthesized — this requires not only knowledge, but also special rare equipment, the distribution of which is monitored.
