Organic chemistry: a complex challenge addressed with AI

IDSIA USI-SUPSI and MeDiTech are working on the Advanced Machine Learning for Innovative Drug Discovery project to support the introduction of AI into the chemical and pharmaceutical industries. The prototype of the algorithm developed to predict the reagents is already being utilized by Pfizer's research division.

Organic chemistry boasts enormous complexity, with tens of billions of compounds currently known. Each compound can also possess a huge number of functional groups that influence the molecule's behavior in specific chemical or biological contexts. Because of this multiplicity of factors, it is very hard to predict the outcome of reactions even when the molecules involved are known.

Computer-aided design is therefore crucial for the pharmaceutical industry, and the use of artificial intelligence offers immense potential for automating these processes. This is particularly evident in the synthesis of drug candidate compounds and in retrosynthesis, which is the search for the steps necessary to obtain a specific compound, such as the active ingredient of a drug, from commercially available components.

The Dalle Molle Institute for Artificial Intelligence (IDSIA USI-SUPSI), in collaboration with the Institute of Digital Technologies for Personalized Healthcare (MeDiTech), is working on the Advanced Machine Learning for Innovative Drug Discovery project to develop innovative methods in which artificial intelligence can provide help to chemists to solve these tasks.

One of the first outcomes concerns the conditions in which a reaction takes place, specifically, the chemical environment. Indeed, organic reactions generally occur in liquid solutions and require specific reactants to proceed. It is common that only through the use of catalysts, solvents or other specific additives can a particular chemical reaction be facilitated, increasing its speed or yield for large-scale production. Predictive methods for identifying appropriate reagents are therefore crucial to automating the planning of chemical synthesis.

"A prototype of our algorithm for predicting reactants that is already being used by the research division of Pfizer, one of our international pharmaceutical partners," says Michael Wand, Senior Researcher SUPSI at IDSIA. "We hope that our methods in the future will help develop new treatments for diseases, and promote a better understanding of the structure of organic chemistry, the basis of all our existence."

Another goal concerns the efficient management of the vast amount of data on chemical reactions  accumulated over decades. Tools to extract insightful information from this huge amount of data would facilitate to better plan chemical processes.
 

References: