Development of machine learning algorithms and an IT system based on them to suggest chemical reaction conditions for key reaction classes

Goal

The goal of the project is to create a novel product that allows planning chemical synthesis pathways taking into account reaction conditions, such as catalyst and temperature.

The work will focus on developing neural networks and machine learning algorithms for suggesting effective conditions for chemical reactions to maximize their efficiency. The plan is to create reaction prioritization algorithms, develop the architecture and methods for training the models, and develop software to use the models in synthesis planning. This will result in a high number of correctly suggested conditions (reaction yields of at least 10%).

The project involves the development of experimental protocols and the execution of 800,000 chemical reactions in 8 key reaction classes. Performing such a large number of reactions will be made possible by using a highly automated chemical laboratory. The project is based on advances in the fields of artificial intelligence and automation, and its uniqueness lies in the combination of these methods, the number of reactions and the level of automation of the laboratory on the scale of the CEE region.

The solution will be enhanced with experimental data and integrated with our existing commercial synthesis planning product. We were the first to launch a synthesis planning product based on deep neural networks in 2019. We plan to enhance it with the largest publicly available dataset from the scientific literature from CAS (USA). Our collection will be a key addition to the available public data due to the presence of negative results, the diversity of chemical structures and the regularity of the collection.

The product innovation will target:

• Large pharmaceutical/biotech companies with internal divisions responsible for drug discovery (DD) processes – about 70% of the market
• Small/medium companies with in-house DD divisions of less than 100 research lines – 20% of the market
• Small/medium companies providing services in the DD sector – 10% of the market

Project budget: 20 968 771.16 PLN

Funding: 14 199 877.80 PLN