AI scientists rapidly design 92 COVID drug candidates

A team of AI-driven virtual scientists identified dozens of potential COVID-19 treatments in just a few days, streamlining a process that would normally take human researchers weeks. Working inside a digital lab, they conducted multiple simultaneous meetings, some lasting only seconds, to plan and execute research strategies.

The project, led by Professor James Cho of Stanford University and Researcher John Park Chan of the Chan Zuckerberg Biohub, was published July 29 in Nature. The system rapidly generated therapeutic concepts for COVID-19, a process that typically demands weeks of work by specialists in infectious disease, biology, and clinical research.

The virtual lab begins when a human assigns a task to a lead AI, which then generates and coordinates a team of specialized agents to address the challenge. These systems, built on large language models, can conduct scientific reasoning and make independent decisions.

During the COVID-19 drug development stage, the lead AI assembled agents with expertise in immunology, computational biology, and machine learning. Within seconds, it created an interdisciplinary research team. A separate critical AI was included to detect weaknesses and offer feedback.

The AI agents shared ideas through rapid meetings. Unlike human sessions that involve breaks or delays, these exchanges ended in seconds or minutes. Some participated in multiple meetings at once, requested tools, and analyzed data on their own. Ideas deemed unrealistic or too costly to verify were excluded in advance.

Rather than conventional antibody treatments, the AIs proposed nanobodies, which are smaller fragments found in animals like llamas. Roughly one-quarter the size of human antibodies, nanobodies allow for more accurate structural modeling. The system generated 92 candidate designs, citing their compact size as an advantage.

Two of the nanobodies were synthesized in a lab and shown to bind strongly to the coronavirus. The findings suggest AI could play a critical role in designing broad-spectrum therapies.

The research team said the lab results are being fed back into the virtual system to improve the nanobody designs. They plan to apply the AI lab to other scientific challenges in the future.

이채린 2bottle9@donga.com

_{Images are for reference only.Images and contents gathered automatic from google or 3rd party sources.All rights on the images and contents are with their legal original owners.}