Creating new medicines, agrochemicals, or cosmetics is often a grueling process for chemists. They must analyze countless chemical reactions to determine whether a desired compound can be synthesized.
Scientists at Yale University have developed an AI system called MOSAIC, which accelerates this process. According to a study published in the journal Nature, the system has already successfully identified the exact factors required to synthesize 35 potentially useful compounds. The synthesis of small molecules is the slowest stage in the drug discovery process, and MOSAIC manages to overcome this barrier.
One popular method of using AI in chemistry is the SMILES system, which converts chemical 3D structures into strings of letters, numbers, and punctuation marks. However, MOSAIC’s goal is more ambitious: to study chemical compounds exactly as chemists themselves do.
Using a database of approximately one million chemical reactions and building upon Meta’s Llama model, the researchers created 2,498 individual models. Each of these specializes in a specific type of molecule.
The operating principle of MOSAIC resembles a team of nearly 2,500 niche specialists, where each member has mastered one specific chemical reaction to perfection. This is precisely why the system is much more accurate. It avoids getting lost in general information and provides chemists with such detailed data that it can be applied directly in the laboratory without any additional research. Furthermore, because these models are small in size, they do not require powerful servers and can run efficiently on standard computers.
