Graph Machine Learning for Chemistry: synthesis planning
Synthesis planning (retrosynthesis) is a fundamental problem in chemistry, widely used in drug discovery. The aim of this process is to understand how to produce a given molecule. Namely, having a target molecule, retrosynthesis aims to find a route from this molecule to commercially available or synthetically known building blocks.
To solve this problem, many methods of Computer-Aided Synthesis Planning were developed. In recent years, Deep Learning models, and particularly, Geometric Deep Learning methods, are used to speed up the process and reduce associated costs.
However, modern models still have many weak spots. Some of the problems are low Top-1 accuracy (although Top-3 accuracy is reasonable), missing reaction conditions, the lack of possibility to learn from negative data, the absence of good metrics, etc.
The research focuses on improvement of (Graph) Machine Learning methods in retrosynthesis. Some of the related topics in Geometric Deep Learning, Explainable Machine Learning and Machine Learning for chemistry and drug design are also under research.
This topic is supervised by M.Sc. Nina Lukashina.
For further information or interest in researching in this area, please contact the Supervisor directly.