Recently, most drug discoveries have been selected and synthesised with the suitable molecules needed by knowledgeable humans to come to the safe and efficient medicine we rely on daily. Scientists often employ a retrosynthesis technique to identify the synthesis. The retrosynthesis technique is a method used in organic chemistry to plan the synthesis of complex molecules. It involves working backwards from a desired target molecule to identify the necessary precursor molecules and the sequence of reactions needed to synthesise the target.
The process of sifting through millions of potential chemical reactions can be a challenging and time-consuming task. With the vast number of possible reactions and transformations that can be applied to different functional groups and molecules, it becomes crucial to streamline the search for suitable reactions during the retrosynthetic analysis.
To address this challenge, several computational tools and software programs have been developed to assist chemists in identifying potential reactions. Researchers at The Ohio State University have created an AI framework called G2Retro to generate reactions for any given molecule automatically.
The recent research demonstrated that the framework outperformed existing manual-planning methods by encompassing a vast structure of potential chemical reactions and effectively and efficiently determining the most suitable reactions for synthesising a specific drug molecule.
“We are seriously concerned in AI development to save human lives, and medicine is what we want to focus on,” said Xia Ning, lead author of the study and an associate professor of computer science and engineering at Ohio State.
The objective of using AI is to accelerate the drug design process, and the results revealed that this approach not only offers time and cost savings for researchers but also generates drug candidates with significantly superior properties compared to naturally occurring molecules.
Ning’s team trained G2Retro using a data set comprising 40,000 chemical reactions compiled from 1976 to 2016. The framework utilises graph-based representations of molecules and employs deep neural networks to generate potential reactant structures for their synthesis.
The framework’s remarkable generative capabilities allow it to generate hundreds of novel reaction predictions within a few minutes when provided with a specific molecule, as highlighted by Ning.
Ning chose this method due to this generative AI method G2Retro can supply multiple different synthesis routes and options, as well as a way to rank other possibilities for each molecule. She also emphasised, “This is not going to replace current lab-based experiments, however, it will offer better drug options so experiments can be prioritised and focused much faster.”
To evaluate the efficacy of the AI system, Ning’s team conducted a case study to assess G2Retro’s ability to accurately predict the synthesis of four recently introduced drugs: Mitapivat, a treatment for hemolytic anemia; Tapinarof, used for various skin diseases; Mavacamten, prescribed for systemic heart failure; and Oteseconazole, employed in the treatment of fungal infections in females.
“G2Retro was able to generate the same patented synthesis routes for these medicines correctly and provided alternative synthesis routes that are also feasible and synthetically useful,” Ning said.
According to Ning, G2Retro successfully produced the exact synthesis pathways for these medications and additionally provided alternative synthesis routes that are both viable and valuable for practical use.
Possessing such a dynamic and efficient tool at scientists’ disposal can facilitate the production of more potent drugs at an accelerated rate.
However, while AI, such as G2Retro, may provide an advantage to scientists within the laboratory, Ning underscored the crucial point that the medications generated by G2Retro or any generative AI still require validation. The validation process needs testing the created molecules in animal models first and subsequently progressing to human trials.
