Uses machine learning and reinforcement learning technology
half of the research period
“Development research on biospecific LLM”
Home SK Telecom announced on the 15th that through joint research with SK Biopharmaceuticals, it has discovered an early active substance that can be used to develop intractable cancer treatments using artificial intelligence (AI).
Through this research, the two companies created and selected a large number of binder candidates that can bind to ROR1, a protein that appears on the surface of cancer cells. Subsequent practical laboratory validation confirmed that two of these binders exhibit potential as initial active materials.
Binders are materials designed to bind to specific targets, such as cancer cells. In order to discover new binders, it is necessary to comprehensively consider several conditions, such as whether they bind well to the target material and whether the material structure is stable. ROR1 is a tumor-associated cell surface protein commonly produced in various blood cancers and solid tumors that occur in the form of lumps in organs and tissues, some of which are more tumorous than normal, and has attracted attention in the field of developing targeted anticancer therapies.
In this study, SK Biopharma established a strategy to discover new binding agents based on its experience in new drug development. SK Telecom created a number of new binder candidates using AI technology, analyzed the possibility of combination with ROR1, and selected candidates for laboratory testing.
In particular, research exploring new material structures often doesn’t have enough data for AI to learn from. For this reason, AI methods that rely only on existing data are limited in their ability to explore a wide range of candidates.
In response, SK Telecom applied machine learning to its research, which expresses protein fragments by combining them in various ways. Additionally, using reinforcement learning (RL), AI can now reward combinations with high structural stability and find optimal new binder structures.
During the screening stage, a large number of new binder candidates were processed in parallel, utilizing SK Telecom’s GPU resources. Since then, the AI model has been able to combine ROR1 with each candidate, quickly predicting and analyzing structures that are likely to actually be combined, and effectively narrowing down the targets for laboratory testing.
As a result, SK Telecom and SK Biopharmaceutical completed the research in about five months. This has shortened the initial research period for new drug development by more than 60% using SK BioPharma’s traditional method, which usually takes one to two years, demonstrating that AI can contribute to reducing the time and cost required in the early stages of new drug development.
SK Telecom’s AI Convergence Headquarters Director Cho Dong-young said, “Based on this result, we are also considering expanding the scope of our technical cooperation to the entire bio-AI field, such as developing a bio-specific LLM (Large Language Model) that utilizes our unique AI-based model.”
