[Reporter's View] what 'AI-driven drug discovery' requires

to encourage more pharmaceutical companies to participate

This year’s Nobel Prize in Chemistry was jointly awarded to David Baker, a professor at the University of Washington, Demis Hassabis, CEO of Google DeepMind, and John Jumper, Director at DeepMind.

 

Professor Baker was recognized for his contribution to creating protein design models.

 

The DeepMind team was honored for developing 'AlphaFold,' which reduced the time required for protein structure prediction from years to mere hours using AI.

 

The potential of AI in drug development has once again been recognized following the winning of the Nobel Prize.

 

At the same time, it has alleviated concerns that AI-driven drug discovery is just chasing a mirage.

 

In South Korea, various AI-driven drug discovery projects are being conducted.

 

The chief project is the 'K-MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery)' project.

 

It is a Korean version of the MELLODDY project, in which 10 global pharmaceutical companies such as Amgen, key European universities, and biotech startups participated in 2020.

 

The core of this project is federated learning.

 

This method gathers data from individual pharmaceutical companies and research institutions to train AI models.

 

To maintain data privacy, it encrypts gathered data.

 

The goal is to shorten the lengthy drug development process significantly.

 

In South Korea, eight pharmaceutical companies, including Daewoong Pharmaceutical·Dong Wha Pharm·Samjin Pharmaceutical·Yuhan Corporation·Jeil Pharmaceutical·Hanmi Pharmaceutical·Huons·JW Pharmaceutical, are participating.

 

The project includes five universities and hospitals, such as Seoul National University Hospital, and four research institutes, including the Korea Research Institute of Bioscience and Biotechnology (KRIBB).

 

AI drug development companies Simplex and Aifase are also part of the project.

 

Experts in AI-driven drug development unanimously emphasize the importance of 'high-quality data' for AI training.

 

A large volume of data is not sufficient.

 

They explain that pharmaceutical companies or research institutions must utilize detailed clinical and preclinical data to accelerate drug discovery.

 

The challenge lies in gathering clinical and preclinical data from individual pharmaceutical companies into one.

 

These data are the intellectual property of each company and the culmination of extensive efforts by numerous research and development teams.

 

While federated learning employs heavily encrypted processes to protect this data, companies remain hesitant and concerned about the potential external leakage of their core proprietary information.

 

The same applies to clinical failure data.

 

Experts stress that failure data is just as important as success data for AI training.

 

Some explain that failure data is more effective in teaching AI systems.

 

However, from the perspective of pharmaceutical companies, systematically recording and managing clinical failure data is a cumbersome task.

 

The pharmaceutical companies participating in the K-MELLODDY project likely have overcome these concerns and challenges to unite around meeting the goal of AI-driven drug development.

 

They deserve recognition.

 

However, it is crucial to encourage more companies to participate.

 

Only then can Korean pharmaceutical companies compete with global pharmaceutical companies that train AI systems on larger datasets worldwide.

 

The government's involvement is crucial to encouraging more Korean companies to participate.

 

The Ministry of Health and Welfare (MOHW) and the Ministry of Science and ICT (MSIT) have agreed to support KRW 34.8 billion in the K-MELLODDY project by 2028.

 

Some argue that while the budget is significant, it may be insufficient to attract greater participation from pharmaceutical companies.

 

In addition to establishing the federated learning platform, direct support for participating companies is essential.

 

The current vision, which promises to reduce the considerable time and costs of drug development alone, is insufficient to give incentive.

 

The government must take a more proactive role so that the goal of creating effective AI models for drug discovery can be closer.