A New Approach to Developing Drugs

Researchers are working on a new approach to computer-aided synthesis planning that will bring new changes to pharmaceutical companies by using artificial intelligence to streamline the new medicine discovery process.

FREMONT, CA: Pharmaceutical companies use artificial intelligence to streamline the process of discovering new medicines. Machine learning models can present new molecules with specific properties that will fight certain diseases within a short period. However, there is a significant hurdle holding these systems back: the models suggest new molecular structures that are difficult to produce in a laboratory.

A new approach from a group of researchers constrains a machine learning model so it suggests molecular structures that can be synthesised. This method ensures composing molecules using materials that can be purchased and that the chemical reactions occurring between those materials follow the chemistry laws. When compared to other methods, these new models proposed molecular structures that scored as high or better using popular evaluations but are synthesizable. Moreover, this system takes less than one second to propose a synthetic pathway, while other methods take several minutes. In a search space that includes billions of potential molecules, those time savings add up.

This process reformulated how these models generated new molecular structures. Many of them think about building new molecular structures. Instead, new approaches construct new molecules, one block at a time, reaction by reaction.

Building Blocks: This new model simulated the process of synthesising a molecule to ensure it could be produced. It has a set of viable building blocks, which are widely available chemicals, and a valid chemical reaction list to work with. These chemical reaction templates are made by experts, and controlling these inputs by allowing certain chemical or specific reactions facilitates the researchers in limiting how large the search space can be for a new molecule.

Furthermore, the models use these inputs to build a tree by selecting building blocks and linking them through chemical reactions to create the final molecule. At every step, the molecule becomes more complex as additional chemicals and reactions are added. It outputs the final molecular structure and the tree of chemicals and reactions that would synthesise it.

Rather than directly designing a product molecule, researchers design an action sequence to obtain that molecule. This allows them to ensure the structure's quality, and they input a complete molecular structure and a set of building blocks and chemical reactions, and the model learns to build a tree that synthesises the molecule to train their model. After learning from numerous examples, the model learns to come up with these synthetic pathways on its own. This new work is exciting and could eventually enable a new paradigm for computer-aided synthesis planning. Moreover, it is likely to be a huge inspiration for future research in the area.