Sanofi’s investment will support a Canadian site’s efforts to apply AI to drug production, while Amgen has unveiled the second expansion of its Puerto Rico plant in quick succession.

In its latest earnings report, the Danish drugmaker said sales and profit should shrink less than the company originally anticipated thanks to strong prescription numbers for its new obesity pill.

​Read More

Developing new molecular compounds is crucial to address pressing challenges, from drug discovery to sustainable materials. However, discovering viable new molecules is challenging due to the vastness of the search space. 

In a new paper published in Nature Machine Intelligence titled, “A collaborative constrained graph diffusion model for the generation of realistic synthetic molecules,” researchers from Universitat Rovira i Virgili (URV) have developed an AI tool capable of generating molecules that are guaranteed to comply with the rules of chemistry. The model, named CoCoGraph, operates similarly to generative AI tools for text or images, such as ChatGPT or Dall-E.  

“These models create new content that looks very much like the real thing. Our algorithm does the same, but with molecules,” said Roger Guimerà, PhD, ICREA research professor in the department of chemical engineering at URV and co-corresponding author of the study. He explains that the number of possible chemical molecules could be up to 10⁶⁰ variants, which is more than the number of water molecules in the ocean.  

CoCoGraph uses a diffusion model, a technique common in image generation, which progressively “disorders” a real molecule and trains the system to learn how to reconstruct it. 

Marta Sales-Pardo, PhD, professor in the department of chemical engineering at URV and co-corresponding author of the study, explains that the model begins with a real molecule, breaks the bonds, and then creates new bonds at random. The model then learns to reverse this process to reconstruct coherent structures. 

Notably, CoCoGraph directly incorporates the basic rules of chemistry, such as maintaining the correct number of bonds, to guarantee that generated molecules are chemically valid. The system is also more efficient, uses fewer parameters, and requires less computing power to generate molecules more quickly. 

The research team has compared the performance of CoCoGraph with other state-of-the-art models and analyzed 36 physicochemical properties, such as solubility and structural complexity. For two-thirds of these properties, the CoCoGraph generated molecules are chemically more realistic than those from other models. 

Although the model cannot yet design molecules with a specific function, researchers have identified molecules with properties similar to the drug, paracetamol. They have also explored techniques to partially modify an existing molecule to create new variants with similar characteristics, which are useful for optimizing drugs or developing new materials. 

The next step is to design molecules with specific properties, such as solubility and low toxicity. If successful, the technology could accelerate the discovery of new solutions across pharmacology and materials science in a chemical universe that is still practically unexplored. 

The post CoCoGraph AI Model Generates Molecules that Comply with Rules of Chemistry appeared first on GEN – Genetic Engineering and Biotechnology News.