Spleen tyrosine kinase (Syk) is active in immune cells and contributes to the development of autoimmune diseases, primarily immune thrombocytopenia. Syk inhibitors are used to treat the disease – these are special compounds that suppress the protein’s activity and thus the accompanying symptoms.
However, not all known compounds are effective or safe: for instance, fostamatinib, one of the approved medications, can cause side effects while not always delivering stable results. That’s why there is an active search for new suitable Syk inhibitors. Conventionally, this is done in-lab via manual search and testing of hundreds or even thousands of molecules. This process can take years and is quite resource-heavy – that’s where AI can come to the rescue.
Researchers from ITMO’s Center for AI in Chemistry have presented a new method for identifying Syk inhibitors; it brings together machine learning and molecule generation. In their work, the team used FREED++, an algorithm that can generate new small molecules for specific therapeutic targets. In order to adapt it to the task, the algorithm was trained to predict the biological activity of compounds from a dataset with 3,176 Syk inhibitors from the open-source library ChEMBL. Now, the algorithm can generate chemical compounds with selected properties and search for new Syk inhibitor candidates in mere days.
A schematic of the algorithm's work. Image courtesy of Maria Zavadskaya
Using the model, the researchers have already been able to produce 139 candidate molecules. Based on computational evaluations, these molecules have a high predicted activity and look more promising than existing analogs: they bind to the target more efficiently and potentially cause fewer side effects. In the future, the team is planning to experimentally test the efficiency and safety of the most promising candidates in vitro and in vivo, as well as train the algorithm to search for other important therapeutic targets – for instance, a similar technology is already used to develop new antibiotics and ophthalmic medicines.
“Our method combines QSAR (quantitative structure-activity relationship – Ed.) modelling and reinforcement learning, which makes it possible to generate highly efficient molecules with minimal side effects. In the nearest future, our molecules will undergo lab testing and, hopefully, attract the attention of pharmaceutical companies before eventually contributing to new effective treatments for autoimmune diseases,” comments Maria Zavadskaya, the paper’s first author and a student at ITMO’s ChemBio Cluster.
Maria Zavadskaya. Photo courtesy of the subject
This research project is supported by the national program Priority 2030.
