SK Telecom, SK Biopharm complete hard-to-treat cancer drug research in 5 months using AI

Early-stage lead compounds identified for targeted cancer therapy

Machine learning, reinforcement learning cut research time by 60%

Companies weigh expansion into biotech LLM development

Researchers from SK Telecom and SK Biopharm discuss findings from their AI-based drug discovery research. [SK Telecom]

SK Telecom used AI to complete a joint research project with SK Biopharm on new drugs for hard-to-treat cancers in about five months.

SK Telecom said Wednesday it had identified early-stage lead compounds suitable for developing targeted therapies against hard-to-treat cancers through AI-assisted joint research with SK Biopharm. Using machine learning and reinforcement learning, the companies cut the research period to about five months — down from the conventional one to two years.

Through the research, the two companies generated and screened a large number of binder candidates capable of attaching to ROR1, a protein found on the surface of cancer cells. Laboratory validation subsequently confirmed that two of the binders showed promise as early-stage lead compounds.

A binder is a substance designed to attach to a specific target such as a cancer cell. Identifying a new binder requires examining multiple criteria simultaneously, including how well it binds to the target and whether its molecular structure is stable.

ROR1 is a tumor-associated cell-surface protein overexpressed in several blood cancers and solid tumors. Because it appears at elevated levels in certain cancer types compared with normal tissue, it has drawn significant attention in targeted cancer therapy development.

In the research, SK Biopharm drew on its drug development experience to devise the strategy for discovering new binders, while SK Telecom applied AI technology to generate a large pool of binder candidates, analyze their binding potential with ROR1 and select candidates for laboratory validation.

According to SK Telecom, research into new molecular structures often suffers from insufficient data for AI training. This limits how broadly candidates can be explored when relying solely on existing data.

To address this, SK Telecom applied machine learning that combines and represents protein fragments in diverse ways. It also used reinforcement learning to reward the AI for combinations with high structural stability, guiding it toward optimal new binder structures.

SK Telecom also said it used GPU resources during the screening phase to process multiple new binder candidates in parallel. An AI model then rapidly predicted and analyzed how each candidate could structurally bind with ROR1 and assessed the likelihood of actual binding.

The joint research with SK Biopharm was completed in about five months as a result — cutting the early-stage drug development timeline, which conventionally takes one to two years, by more than 60%. SK Telecom said the achievement demonstrates that AI can dramatically reduce the time and cost involved in the early stages of drug development.

Jo Dong-yeon, head of AI convergence at SK Telecom, said the company is considering expanding its technology cooperation into the broader biotech AI field, including developing a biotech-specialized large language model built on its proprietary AI foundation model.

chami@heraldcorp.com