(Nanowerk News) While it takes the pharmaceutical industry years to create drugs capable of treating or curing human disease, a new study shows that using generative artificial intelligence can speed up the drug development process.
Today, most drug discovery is done by human chemists who rely on their knowledge and experience to select and synthesize the right molecules needed to become the safe and efficient drugs we depend on. To identify synthesis pathways, scientists often use a technique called retrosynthesis – a method of making potential drugs by working backwards from the desired molecule and looking for chemical reactions to make it.
But because sifting through millions of potential chemical reactions can be a very challenging and time-consuming endeavor, researchers at The Ohio State University have created an AI framework called G2Retro to automatically generate reactions for any given molecule. The new study shows that compared to current manual planning methods, this framework is able to cover a wide range of possible chemical reactions and to accurately and quickly distinguish which reactions are most successful for making a particular drug molecule.
“Using AI for things that are important for saving human lives, like pharmaceuticals, is what we really want to focus on,” said Xia Ning, lead author of the study and professor of computer science and engineering at Ohio State. “Our goal is to use AI to speed up the drug design process, and we found that AI not only saves researchers time and money, but also provides drug candidates that may have properties that are far superior to any molecule present in nature.”
The study builds on Ning’s previous research in which his team developed a method called Modof that is able to produce molecular structures that exhibit the desired properties better than existing molecules. “Now the question is how to make the resulting molecule like that, and that’s where this new study shines,” said Ning, also a professor of biomedical informatics in the College of Medicine.
This study is published in the journal Communication Chemistry (“G2Retro as a two-step graph generative model for retrosynthesis prediction”).
Team Ning trains G2Retro on a dataset of 40,000 chemical reactions collected between 1976 and 2016. The framework “learns” from a graphical representation of a given molecule, and uses deep neural networks to generate possible structures for reactants that can be used to synthesize them. Its generative powers are so impressive that, according to Ning, after being given a molecule, G2Retro can generate hundreds of new reaction predictions in just a few minutes.
“Our generative AI method G2Retro can provide several different synthesis routes and options, as well as a way to rank different options for each molecule,” said Ning. “This won’t replace current laboratory-based experiments, but it will offer a wider and better selection of drugs so experiments can be prioritized and focused more quickly.”
To further test the effectiveness of AI, Ning’s team conducted a case study to see if G2Retro can accurately predict four new drugs already in circulation: Mitapivat, a drug used to treat hemolytic anemia; Tapinarof, which is used to treat various skin ailments; Mavacamten, a drug to treat systemic heart failure; and Oteseconazole, used to treat yeast infections in women. G2Retro was able to properly generate a patented synthesis route for these drugs, and provide an alternative synthesis route that is also synthetically feasible and useful, said Ning.
Having dynamic and effective tools available to scientists could allow the industry to produce more powerful drugs at a faster rate – but while the edge of AI can be given to scientists in the lab, Ning emphasizes G drugs2Any retro or generative AI still needs validation – a process that involves creating molecules to be tested in animal models and then in human trials.
“We are excited about generative AI for medicine, and we are dedicated to using AI in a responsible way to improve human health,” said Ning.