back to top

    Artificial intelligence to teach robots how to quickly produce drugs

    Press a button to produce any of thousands of organic chemicals, such as drugs, in a few moments from readily available ingredients. Why not? The next step in this direction has just been made possible by a peculiar collaboration between artificial intelligence (Polish program) and robots performing chemical reactions.

    The teams of Prof. Martin Burke and Prof. Bartosz Grzybowski describe in “Science” how to further automate the synthesis of an important group of chemical reactions. Specifically, these are Suzuki-Miyaura coupling. Such reactions (Nobel Prize 2010) result in the formation of carbon-carbon bonds. “This type of reaction allows blocks of chemical compounds to be put together as easily as putting Lego blocks together. This is now the basis of medicinal chemistry. We are able to assemble most of the known anticancer drugs from the currently known building blocks,” Prof. Grzybowski explains.


    The teams have identified general conditions under which such Suzuki reactions can be carried out.


    “This work is an exciting step toward a world where anyone can make molecules,” Prof. Martin Burke of the University of Illinois comments for Science in Poland.


    A while back, Prof. Martin Burke and his team developed robots that build predetermined compounds from available chemical building blocks. They carry out chemical reactions and even form complex organic compounds (which are based on chains or carbon rings). However, these robots were still only on the road to “independence.” After all, before carrying out any new reaction, it was the human who had to indicate under what conditions the reaction should take place – that is, for example, what temperature is needed. 


    Extremely innovative was the way the team searched for universal reaction conditions. Robots and artificial intelligence were harnessed to work together.


    “We used an artificial intelligence module designed by my team, Allchemy, which predicts reaction conditions – that is, it suggests how to choose catalysts or solvent to produce a certain product from given substrates,” Prof. Bartosz Grzybowski explains.


    The work involved testing a very different set of reaction conditions and finding the best ones. The artificial intelligence indicated to the robots what reactions they should perform and under what conditions they needed to do so. The robots carried out its recommendations and gave feedback on how the reaction went. On the basis of this closed-loop process, the artificial intelligence built itself the knowledge base necessary to show the robots the way to the best possible set of conditions in subsequent steps.


    The robots made more than 500 such responses. In this way, artificial intelligence has developed a pretty good idea of which reaction conditions are favourable for very different systems.


    While the efficiency of chemical reactions conceived traditionally by top specialists – was 25 per cent, artificial intelligence in partnership with robots reached an efficiency of more than 50 per cent. This is a big difference, reports Prof. Grzybowski.


    Asked whether the solution will enable robots to produce various compounds, such as anti-cancer drugs, on demand, Prof. Grzybowski says this is one of the intentions of a project he is leading – which has funding from DARPA, an agency of the US military.


    Check the article here


    More in section