From decades to a second: AI and supercomputer reshape drug discovery

A new AI‑driven platform from Tsinghua University, paired with China's exascale supercomputer Tianhe, has slashed drug‑screening time from decades to under a second per target.

The system, called Galaxy VS, hosts over 100 billion virtual molecules and millions of protein targets from humans and other species. It is open to researchers worldwide – for free.

The team treated protein structures as "search terms" and trained AI to match drug candidates in milliseconds – like a search engine for molecules.

Traditional methods would take decades if not centuries to test one target against a 100‑million‑molecule library using a single CPU core, according to a 2020 Harvard Medical School estimate.

The computational heavy lifting is done by Tianhe, China's next‑generation exascale supercomputer in Tianjin. In one second, it completes a calculation that would take all 8 billion people on Earth about four years to finish.

But raw power wasn't enough. Engineers spent over 700 days translating and optimizing Tsinghua's AI code into "MilkyWay language" – testing hundreds of thousands of lines line by line.

The result? "We estimated a few minutes per target," said Dr. Li Peishun, senior engineer at the center. "But when we ran the full screening against 100 billion molecules, each target took less than one second."

Early validation with top global drugmakers shows the AI predictions match lab results with high accuracy.

While clinical translation remains the real test, Galaxy VS removes the biggest early bottleneck. In an industry where 90% of candidates fail, failing faster may be the quickest path to success.