SK Telecom and SK Biopharmaceuticals used artificial intelligence (AI) to identify early hit compounds that can be applied to develop targeted therapies for intractable cancers. They also shortened the early research phase of new drug development, which typically took 1 to 2 years, to about five months.
SK Telecom said on the 15th that, through joint research with SK Biopharmaceuticals, it designed and analyzed binder candidates that bind to "ROR1," a cell-surface protein in cancer cells, and confirmed the potential of two of them through experiments. A binder is a substance designed to selectively bind to a specific disease target.
ROR1 is a protein that is expressed at higher levels in some blood cancers and solid tumors than in normal cells, drawing attention in the field of anticancer targeted therapy development. Based on its experience in new drug development, SK Biopharmaceuticals established a strategy for discovering candidate substances and a verification system, while SK Telecom used AI to generate a large number of new molecular structures and narrowed down candidates with a high likelihood of binding.
Considering the characteristics of the new drug development field, where training data are scarce, the researchers applied a Machine Learning technique that combines protein fragments, or fragments, in various forms. Through Reinforcement Learning, they designed the system to give higher rewards to structurally stable combinations so that the AI would search for the optimal binder structure.
SK Telecom's graphics processing unit (GPU) resources were used in the candidate selection process. After simultaneously analyzing numerous substances, the AI model predicted the binding structures and likelihood with ROR1 to narrow down the actual experimental targets. The company said this reduced the early research period by more than 60% compared with conventional methods.
Building on this achievement, SK Telecom plans to consider expanding the scope of collaboration with SK Biopharmaceuticals, including developing a bio-specialized large language model using its own AI foundation model.